A Macroeconomic Model of Equities and Real, Nominal, and Defaultable Debt Eric T. Swanson University of California, Irvine eric.swanson @ uci.edu http://www.ericswanson.org Abstract I show that a simple, fully structural New Keynesian model with Epstein-Zin preferences is consistent with a wide variety of asset pricing facts, such as the size and variability of risk premia on equities, real and nominal government bonds, and corporate bonds—the equity premium puzzle, bond premium puzzle, and credit spread puzzle, respectively. I thus show how to unify a variety of asset pricing puzzles from finance into a simple, structural framework. Conversely, I show how to bring standard macroeconomic models into agreement with a wide range of asset pricing facts. JEL Classification: E32, E43, E44, E52, G12 Version 2.0 February 28, 2019 I thank Martin Andreasen, Ian Dew-Becker, Anthony Diercks, Leland Farmer, Andrew Glover, SungJun Huh, Sebasti´ an Infante, Hanno Lustig, Jean-Paul Renne, Anastasia Zervou, and semi- nar participants at the Aarhus/CREATES Macro-Finance workshop, NBERSI DSGE Modeling Meeting, Macro Modeling Workshop at the Banca d’Italia, EABCN-ECB-FRB Atlanta Nonlin- earities in Macro Conference, SED Meetings, Econometric Society World Congress, UT Austin, Bank of Canada/Federal Reserve Bank of San Francisco Conference on Fixed Income, and Santi- ago Finance Workshop for helpful discussions, comments, and suggestions. The views expressed in this paper, and any errors and omissions, should be regarded as those solely of the author and are not necessarily those of the individuals or groups listed above.
57
Embed
A Macroeconomic Model of Equities and Real, Nominal, and …swanson2/papers/ezap.pdf · 2019. 2. 28. · The model has two essential ingredients: generalized recursive preferences,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
I show that a simple, fully structural New Keynesian model with Epstein-Zinpreferences is consistent with a wide variety of asset pricing facts, such as thesize and variability of risk premia on equities, real and nominal governmentbonds, and corporate bonds—the equity premium puzzle, bond premium puzzle,and credit spread puzzle, respectively. I thus show how to unify a variety of assetpricing puzzles from finance into a simple, structural framework. Conversely, Ishow how to bring standard macroeconomic models into agreement with a widerange of asset pricing facts.
JEL Classification: E32, E43, E44, E52, G12
Version 2.0
February 28, 2019
I thank Martin Andreasen, Ian Dew-Becker, Anthony Diercks, Leland Farmer, Andrew Glover,SungJun Huh, Sebastian Infante, Hanno Lustig, Jean-Paul Renne, Anastasia Zervou, and semi-nar participants at the Aarhus/CREATES Macro-Finance workshop, NBERSI DSGE ModelingMeeting, Macro Modeling Workshop at the Banca d’Italia, EABCN-ECB-FRB Atlanta Nonlin-earities in Macro Conference, SED Meetings, Econometric Society World Congress, UT Austin,Bank of Canada/Federal Reserve Bank of San Francisco Conference on Fixed Income, and Santi-ago Finance Workshop for helpful discussions, comments, and suggestions. The views expressedin this paper, and any errors and omissions, should be regarded as those solely of the author andare not necessarily those of the individuals or groups listed above.
1
1. Introduction
Traditional macroeconomic models, such as Christiano, Eichenbaum, and Evans (2005) and Smets
and Wouters (2007), ignore asset prices and risk premia and, in fact, do a notoriously poor job of
matching financial market variables (e.g., Mehra and Prescott, 1985; Backus, Gregory, and Zin,
1989; Rudebusch and Swanson, 2008). At the same time, traditional finance models, such as Dai
and Singleton (2003) and Fama and French (2015), ignore the real economy; even when these
models use a stochastic discount factor or consumption rather than a latent factor framework,
those economic variables are taken to be exogenous, reduced-form processes.
Despite this traditional separation, the linkages between the real economy and financial
markets can be very important. During the 2007–09 global financial crisis and 2010–14 Eu-
ropean sovereign debt crisis, concerns about asset values caused lending and the real economy
to plummet, while the deteriorating economy caused private-sector risk premia to increase and
asset prices to spiral further downward (e.g., Mishkin, 2011; Gorton and Metrick, 2012; Lane,
2012). These crises also led to dramatic fiscal and monetary policy interventions that were well
beyond the range of past experience.1 Reduced-form finance models that perform well based on
past empirical correlations may perform very poorly when those past correlations no longer hold,
such as when there is a structural break or unprecedented policy intervention as observed during
these crises. A structural macroeconomic model is more robust to these types of breaks and can
immediately provide insights into their effects on risk premia, financial markets, and the real
economy. Macroeconomic models can also provide useful intuition about why output, inflation,
and asset prices co-move in certain ways and how that comovement may change in response to
policy interventions and structural breaks.
In the present paper, I develop a simple, structural macroeconomic model that is consistent
with a wide range of asset pricing facts, such as the size and variability of risk premia on equities
and real, nominal, and defaultable debt. Thus, unlike traditional macroeconomic models, the
model I present here is able to match asset prices and risk premia remarkably well. Unlike tradi-
tional finance models, the model I develop here can provide insight into the effects of novel policy
1For example, the U.S. Treasury bought large equity stakes in automakers and financial institutions, and insuredmoney market mutual funds to prevent them from “breaking the buck.” The Federal Reserve purchased very largequantities of longer-term Treasury and mortgage-backed securities and gave explicit forward guidance about thelikely path of the federal funds rate for years into the future. See, e.g., Mishkin (2011) and Gorton and Metrick(2012). For Europe, the European Union established the European Stability Mechanism to provide quick financialbacking to member countries in need, while the European Central Bank provided large and unprecedented three-year loans to banks and announced that it would purchase large quantities of euro area members’ bonds if theyields on those bonds became excessively stressed (e.g., Wall Street Journal, 2012; European Central Bank, 2012).
2
interventions and structural breaks on asset prices, and gives a unified structural explanation for
the behavior of risk premia on a variety of assets.
The model has two essential ingredients: generalized recursive preferences, as in Epstein
and Zin (1989), Weil (1989), and Tallarini (2001), and nominal rigidities, as in the textbook
New Keynesian models of Woodford (2003) and Galı (2008). Generalized recursive preferences
allow the model to generate substantial risk premia without greatly distorting the behavior of
macroeconomic aggregates, while nominal rigidities allow the model to match the behavior of
inflation, nominal interest rates, and nominal assets such as Treasuries and corporate bonds.
My results have important implications for both macroeconomics and finance. For macroe-
conomics, I show how standard dynamic structural general equilibrium (DSGE) models can be
brought into agreement with a wide variety of asset pricing facts. I thus address Cochrane’s
(2008) critique that a failure of macroeconomic models to match even basic asset pricing facts
is a sign of fundamental flaws in those models.2 Moreover, bringing those models into better
agreement with asset prices makes it possible to use those models to study the linkages between
risk premia, financial markets, and the real economy.
For finance, I unify a variety of asset pricing puzzles into a simple, structural framework.
This framework can then be used to study the relationships between the different puzzles with
each other and with the economy. For example, Backus, Gregory, and Zin (1989), Donaldson,
Johnsen, and Mehra (1990), and Den Haan (1995) argue that the yield curve ought to slope
downward on average because interest rates tend to be low during recessions, implying that bond
prices are high when consumption is low (which would lead to an insurance-like, negative risk
premium). According to the model here, the nominal yield curve can slope upward even if the real
yield curve slopes downward if technology shocks (or other “supply” shocks) are an important
source of economic fluctuations. Technology shocks cause inflation to rise when consumption
falls, so that long-term nominal bonds lose rather than gain value in recessions, implying a
positive risk premium. Similarly, the model developed here can be used to study the changes
in correlations between stock and bond returns documented by Baele, Bekaert, and Inghelbrecht
(2010), Campbell, Sundaram, and Viceira (2013), and others.
Previous macroeconomic models of asset prices have tended to focus exclusively on a single
2As Cochrane (2008) points out, asset markets are the mechansim by which marginal rates of substitution areequated to marginal rates of transformation in a macroeconomic model. If the model is wildly inconsistent withbasic asset pricing facts, then by what mechanism does the model equate these marginal rates of substitution andtransformation?
3
type of asset, such as equities (e.g., Boldrin, Christiano, and Fisher, 2001; Tallarini, 2001; Gu-
venen, 2009; Barillas, Hansen, and Sargent, 2009) or debt (e.g., Rudebusch and Swanson, 2008,
2012; Van Binsbergen et al., 2012; Andreasen, 2012). A disadvantage of this approach is that it’s
unclear whether the results in each case generalize to other asset classes. For example, Boldrin,
Christiano, and Fisher (2001) show that capital immobility in a two-sector DSGE model can fit
the equity premium by making the price of capital (and equity) more volatile, but this mecha-
nism does not explain subtantial risk premia on long-term government bonds, which involve the
valuation of a fixed stream of coupon payments. By focusing on multiple asset classes, I impose
additional discipline on the model and ensure that its results apply more generally. Matching the
behavior of a variety of assets also helps to identify model parameters, since different types of
assets are relatively more informative about different aspects of the model. For example, nominal
assets are helpful for identifying parameters related to inflation, and long-lived equities provide
information about the longer-run features of the model.
The three papers most closely related to the present paper are Tallarini (2001), Rudebusch
and Swanson (2012), and Campbell, Pflueger, and Viceira (2018). Tallarini (2001) incorporates
Epstin-Zin-Weil preferences into a real business cycle model to match the equity premium. Rela-
tive to Tallarini (2001), the model here matches nominal as well as real features of the economy;
explains the behavior of multiple assets, such as equities and real, nominal, and defaultable debt;
and works within the now-standard New Keynesian DSGE framework rather than a real business
cycle framework, which allows monetary policy to have nontrivial effects on asset prices and the
economy. Following Tallarini (2001), Rudebusch and Swanson (2012) incorporate Epstein-Zin-
Weil preferences into a standard New Keynesian DSGE model to match the behavior of nominal
long-term bonds.3 In contrast to Rudebusch and Swanson (2012), the model here is substantially
simpler (to clarify its essential features) and is extended to match the behavior of multiple asset
classes. Campbell, Pflueger, and Viceira (2018, henceforth CPV) study stock and bond prices in
a reduced-form three-equation New Keynesian model. In contrast to the present paper, CPV use
a stochastic discount factor that has an exogenous, reduced-form relationship to output in their
model.4 In fact, the exogenous mapping from output to the stochastic discount factor is crucial
3See also Van Binsbergen et al. (2012) and Andreasen (2012) for variations on the analysis in Rudebusch andSwanson (2012).
4 In this respect, their analysis is similar to the term-structure studies of Rudebusch and Wu (2007) and Bekaert,Cho, and Moreno (2010), both of whom also use a reduced-form three-equation New Keynesian framework and astochastic discount factor that is related to the economy in a reduced-form manner.
4
for CPV’s results: as shown by Lettau and Uhlig (2000) and Rudebusch and Swanson (2008),
CPV’s Campbell-Cochrane (1999) habit specification is typically unable to produce significant
risk premia in a structural model when households can endogenously smooth consumption, be-
cause households endogenously choose a path for consumption that is so smooth the stochastic
discount factor is stabilized.5 In the present paper, I undertake a more structural approach, spec-
ifying a complete—but simple—macroeconomic model in which the stochastic discount factor is
internally consistent with the rest of the model.
A number of other recent papers have begun to study stock and bond prices jointly in a
traditional affine framework (e.g., Eraker, 2008; Bekaert, Engstrom, and Grenadier, 2010; Lettau
and Wachter, 2011; Ang and Ulrich, 2013; Koijen, Lustig, and Van Nieuwerburgh, 2013).6 Some
of these studies work with latent factors, ignoring the real economy, while others relate asset
prices to the reduced-form behavior of consumption. However, none of them uses a structural
macroeconomic model, which has the advantages described above. Although reduced-form models
often fit the data better than structural macroeconomic models, this can simply be a tautological
implication of Roll’s (1977) critique (that any mean-variance efficient portfolio perfectly fits the
mean returns of all assets), as noted by Cochrane (2008). It is only the correspondence of financial
risk factors to plausible economic risks that makes reduced-form financial factors interesting.
Chen, Collin-Dufresne, and Goldstein (2009), Bhamra, Kuehn, and Strebulaev (2010), and
Chen (2010) model equity and corporate bond prices jointly in an endowment economy. Those
authors undertake a more detailed, structural analysis of the corporate financing decision than I
consider here, but at a cost of working in a much simpler, reduced-form macroeconomic environ-
ment. In other words, I use a simple, reduced-form model of the firm in order to better focus on
the structural behavior of the economy, while Chen et al. (2010), Bhamra et al. (2010), and Chen
(2010) use a simple, reduced-form model of the macroeconomy to better focus on the structural
finance behavior of the firm. The advantages of the structural macroeconomic approach I take
here are discussed above.
Throughout the present paper, a recurring theme is the simplicity of the model, in the
5Households with Campbell-Cochrane (1999) habits are extremely averse to high-frequency fluctuations inconsumption. In a DSGE model (as opposed to an endowment economy), households can self-insure themselvesfrom these fluctuations by varying their hours of work or savings. In fact, for plausible parameterizations of DSGEmodels, households endogenously choose a path for consumption that is so smooth the stochastic discount factordoes not vary much more than in the model without habits, leading risk premia to be about the same as withouthabits. See Rudebusch and Swanson (2008) and Lettau and Uhlig (2000).
6See also Campbell, Sundaram, and Viceira (2012), who price stocks and bonds jointly in a quadratic latent-factor framework.
5
interest of clarity and to help provide intuition for the underlying mechanisms. Thus, the model
here is not designed to match very detailed features of the economy or asset prices; indeed, if
one pushes the model far enough, it is certain to fail at matching some features of financial
markets or the macroeconomy. (In a way, that failure is by design, due to the model’s emphasis
on simplicity.) Thus, the model here should be viewed as a “proof of concept” that the standard
New Keynesian DSGE framework can be adapted to match asset prices quite well and shows a
great deal of promise for future development in this direction. The approach I take here is thus
analogous to Kydland and Prescott (1982), who showed that the stochastic growth model could
be extended to match key features of business cycle fluctuations. Their stylized model failed to
match many details of business cycles (e.g., unemployment, inflation), but opened the door to
the equilibrium modeling of these phenomena.
The remainder of the paper proceeds as follows. In Section 2, I develop a simple New
Keynesian DSGE model with Epstein-Zin preferences, show how to solve the model, and discuss
the calibration of the model and its implications for macroeconomic quantities. In Section 3,
I derive the prices of stocks and real, nominal, and defaultable bonds within the framework of
the model, and compare the behavior of those asset prices to the data. In Section 4, I discuss a
number of important features of the model, such as endogenous conditional heteroskedasticity and
endogenous uncertainty vs. exogenous uncertainty shocks. Section 6 concludes. Two Appendices
present all the equations of the model, discuss the numerical solution method in more detail, and
provide additional figures and analysis of the basic results.
2. A Simple Macroeconomic Model
In this section, I develop a simple dynamic macroeconomic model with generalized recursive
preferences and nominal rigidities. Generalized recursive preferences, as in Epstein and Zin (1989)
and Weil (1989), are required for the model to match the size of risk premia in the data.7 Nominal
rigidities are necessary for the model to match the basic behavior of inflation, nominal interest
rates, and nominal assets such as Treasuries and corporate bonds.
Throughout this section, I strive to keep the model as simple as possible while still matching
the essential behavior of macroeconomic variables and asset prices. The goal is to maximize
intuition and insight into the relationships between the macroeconomy and asset prices, and
7See the previous footnote and Rudebusch and Swanson (2008) for a discussion of why habits in householdpreferences, such as Campbell and Cochrane (1999), are unable to match the size of risk premia in DSGE models.
6
avoid tangential complications. For this reason, I deliberately follow the very simple, “textbook”
New Keynesian models of Woodford (2003) and Galı (2008), extended to the case of generalized
recursive preferences. In principle, more realistic, medium-scale New Keynesian models such as
Christiano, Eichenbaum, and Evans (2005) and Smets and Wouters (2007) could also be extended
to the case of Epstein-Zin preferences to achieve an even better empirical fit to the data, but at
the cost of being much more complicated.
2.1 Households
Time is discrete and continues forever. There is a unit continuum of representative households,
each with generalized recursive preferences as in Epstein and Zin (1989) and Weil (1989). In each
period t, the representative household receives the utility flow
u(ct, lt) ≡ log ct − ηl1+χt
1 + χ, (1)
where ct and lt denote household consumption and labor in period t, respectively, and η > 0
and χ > 0 are parameters. Note that equation (1) differs from Epstein and Zin (1989) and Weil
(1989) in that period utility depends on labor as well as consumption.
The assumption of additive separability in (1) follows Woodford (2003) and Galı (2008) and
simplifies many aspects of the model. For example, the household’s intertemporal elasticity of
substitution is unity, its Frisch elasticity of labor supply is 1/χ, and its stochastic discount factor
(defined below) is related to ct+1/ct; without additive separability, the expressions for these
quantities would all be much more complicated. The similarity of the stochastic discount factor
to versions of the model without labor also facilitates comparison to the finance literature. In
addition, assuming logarithmic preferences over consumption ensures that the model is consistent
with balanced growth (King, Plosser, and Rebelo, 1988, 2002) and is a standard benchmark in
macroeconomics (e.g., King and Rebelo, 1999).
Households can borrow and lend in a default-free one-period nominal bond market at the
continuously-compounded interest rate it. The use of continuous compounding simplifies the
bond-pricing equations below and enhances comparability to the finance literature. Each period,
the household faces a flow budget constraint
at+1 = eitat + wtlt + dt − Ptct, (2)
7
where at denotes beginning-of-period nominal assets, wt and dt denote the nominal wage and
exogenous transfers to the household, respectively, and Pt denotes the price of consumption. The
household faces a standard no-Ponzi-scheme constraint,
limT→∞
Et
T∏τ=t
e−iτ+1aT+1 ≥ 0. (3)
Let (ct, lt) ≡ {(cτ , lτ )}∞τ=t denote a state-contingent plan for household consumption and
labor from time t onward, where the explicit state-dependence of the plan is suppressed to reduce
notation. Following Epstein and Zin (1989), Weil (1989), and Tallarini (2001), I assume that the
household has preferences over state-contingent plans ordered by the recursive functional
V (ct, lt) = (1− β) u(ct, lt) − βα−1log[Et exp
(−αV (ct+1, lt+1))], (4)
where β ∈ (0, 1) and α ∈ R are parameters, Et denotes the mathematical expectation conditional
on the state of the economy at time t, and (ct+1, lt+1) denotes the state-contingent plan (ct, lt)
from date t + 1 onward. The case α = 0 in (4) is defined by letting α → 0 and corresponds to
the special case of expected utility preferences. When α �= 0 in (4), the expectation operator is
effectively “twisted” and “untwisted” by the exponential function with coefficient −α. This leavesthe household’s intertemporal elasticity of substitution in (4) the same as for expected utility, but
amplifies (or attenuates) the household’s risk aversion with respect to gambles over future utility
flows by the additional curvature parameter α, with larger values of α corresponding to greater risk
aversion. Thus, generalized recursive preferences allow the household’s intertemporal elasticity of
substitution and coefficient of relative risk aversion to be parameterized independently. Following
Hansen and Sargent (2001), the specific form of generalized recursive preferences in (4) is often
referred to as “multiplier preferences”.
In each period, the household maximizes (4) subject to the budget constraint (2)–(3). The
state variables of the household’s optimization problem are at and Θt, where the latter is a
vector denoting the state of the aggregate economy at time t. The household’s “generalized value
function” V (at; Θt) satisfies the generalized Bellman equation
V (at; Θt) = max(ct,lt)
(1− β) u(ct, lt) − βα−1log[Et exp
(−αV (at+1; Θt+1))], (5)
where at+1 is given by (2).
8
It’s straightforward to show (e.g., Rudebusch and Swanson, 2012), that the household’s
stochastic discount factor is given by
mt+1 ≡ βctct+1
exp(−αV (at+1; Θt+1)
)Et exp
(−αV (at+1; Θt+1)) . (6)
Let rt denote the one-period continuosly-compounded risk-free real interest rate. Then
e−rt = Etmt+1. (7)
2.2 Firms
The economy also contains a continuum of infintely-lived monopolistically competitive firms in-
dexed by f ∈ [0, 1], each producing a single differentiated good. Firms hire labor from households
in a competitive market and have identical Cobb-Douglas production functions,
yt(f) = Atk1−θlt(f)
θ, (8)
where yt(f) denotes firm f ’s output, At is aggregate productivity affecting all firms, k and lt(f)
denote the firm’s capital and labor inputs at time t, respectively, and θ ∈ (0, 1) is a parameter.
For simplicity, and following Woodford (2003) and Galı (2008), I assume that firms’ capital stocks
are fixed, so that labor is the only variable input to production. Intuitively, movements in the
capital stock are small at business-cycle frequencies and are dominated by fluctuations in labor.8
Technology, At, follows an exogenous AR(1) process,
logAt = ρA logAt−1 + εAt , (9)
where ρA ∈ (−1, 1], and εAt denotes an i.i.d. white noise process with mean zero and variance σ2A.
For simplicity and comparability to the finance literature, I set ρA = 1 in the baseline calibration
of the model, below, but consider alternative values of ρA as well. For simplicity and ease of
exposition, I abstract from technology growth in the baseline calibration of (9) as well (i.e., the
mean of log(At/At−1) is 0).9
8Woodford (2003, p. 167) compares a model with fixed firm-specific capital to a model with endogenous capitaland investment adjustment costs and finds that the basic business-cycle features of the two models are very similar.In models with endogenous capital (e.g., Christiano et al., 2005; Smets and Wouters, 2007; Altig et al., 2011),investment adjustment costs are typically included to keep the capital stock stable at higher frequencies. Thus,one can think of the fixed-capital assumption as a simple way of achieving the same result. Woodford (2003) andAltig et al. (2011) also show that firm-specific capital stocks help generate inflation persistence that is consistentwith the data (see particularly Woodford, 2003, pp. 163-173).
9 If the mean rate of technology growth is μA, then the firm-specific capital stocks k must also grow at rateμA/θ in order for the model to have balanced growth.
9
Firms set prices optimally subject to nominal rigidities in the form of Calvo (1983) price
contracts, which expire with exogenous probability 1 − ξ each period, ξ ∈ [ 0, 1). Each time a
Calvo contract expires, the firm sets a new contract price p∗t (f) freely, which then remains in effect
for the life of the new contract, with indexation to the (continuously-compounded) steady-state
inflation rate π each period.10 In each period τ ≥ t that the contract remains in force, the firm
must supply whatever output is demanded at the contract price p∗t (f)e(τ−t)π, hiring labor lτ (f)
from households at the market wage wτ .
Firms are jointly owned by households and distribute all profits and losses back to house-
holds each period in an aliquot, lump-sum manner. When a firm’s price contract expires, the
firm chooses the new contract price p∗t (f) to maximize the value to shareholders of the firm’s cash
flows over the lifetime of the contract,11
∞∑j=0
ξjEt
{mt,t+j(Pt/Pt+j)
[pt+j(f)yt+j(f)− wt+j lt+j(f)
] ∣∣∣ pt+j(f)= p∗t (f)ejπ}, (10)
where mt,t+j ≡∏j
i=1mt+i denotes shareholders’ stochastic discount factor from period t+ j back
to t, Pt the aggregate price level (defined below), wt the nominal wage at time t, and yt+j(f)
and lt+j(f) denote the firm’s output and labor in period t + j, respectively, conditional on the
contract price p∗t (f) still being in effect.
The output of each firm f is purchased by a perfectly competitive final goods sector, which
aggregates the differentiated goods into a single final good using a CES production technology,
Yt =
[∫ 1
0
yt(f)1/λdf
]λ, (11)
where Yt denotes the quantity of the final good and λ > 1 is a parameter. Each intermediate firm
f thus faces a downward-sloping demand curve for its product with elasticity λ/(1− λ),
yt(f) =
(pt(f)
Pt
)−λ/(λ−1)
Yt, (12)
where pt(f) denotes the price in effect for firm f at time t (so pt(f) = p∗τ (f)e(t−τ)π, letting
τ ≤ t denote the most recent period in which firm f reset its contract price), and Pt is the CES
10The assumption of indexation keeps the model well-behaved with respect to changes in steady-state inflation.The continuous compounding is notationally simpler for some of the equations below.
11Equivalently, the firm can be viewed as choosing a state-contingent plan for prices that maximizes the valueof the firm to shareholders.
10
aggregate price of the final good,
Pt ≡[∫ 1
0
pt(f)1/(1−λ)df
]1−λ
. (13)
Differentiating (10) with respect to p∗t (f) and setting the derivative equal to zero yields the
standard New Keynesian price optimality condition,
p∗t (f) = λ
∑∞j=0 ξ
jEt
{mt,t+j(Pt/Pt+j)yt+j(f)μt+j(f)
∣∣ pt+j(f)= p∗t (f)ejπ}∑∞
j=0 ξjEt
{mt,t+j(Pt/Pt+j)yt+j(f)ejπ
∣∣ pt+j(f)= p∗t (f)ejπ} , (14)
where μt(f) denotes the (nominal) marginal cost for firm f at time t,
μt(f) ≡ wtlt(f)
θyt(f). (15)
That is, the firm’s optimal contract price p∗t (f) is a monopolistic markup λ over a discounted
weighted average of expected future marginal costs over the lifetime of the contract.12
2.3 Aggregate Resource Constraints and Government
Let Lt denote the aggregate quantity of labor demanded by firms,
Lt =
∫ 1
0
lt(f)df. (16)
Then Lt satisfies
Yt = Δ−1t AtK
1−θLθt , (17)
where K = k denotes the aggregate capital stock and
Δt ≡[∫ 1
0
(pt(f)
Pt
)λ/((1−λ)θ)
df
]θ
(18)
measures the cross-sectional dispersion of prices across firms. Δt has a minimum value of unity
when pt(f) = Pt for all firms f ; a greater degree of cross-sectional price dispersion increases Δt
and reduces the economy’s efficiency at producing final output.13
Labor market equilibrium requires that Lt = lt, firms’ labor demand equals the aggregate
labor supplied by households. Equilibrium in the final goods market requires Yt = Ct, where
12To be more precise, p∗t (f) is a weighted average of marginal costs deflated by the inflation index rate,
μt+j(f)/ejπ . In addition, the weights in (14) depend on yt+j(f), which depend on the left-hand-side variable
p∗t (f), so (14) is not a closed-form solution for p∗t (f). However, the closed-form solution for p∗t (f), reported in theAppendix, has the same form as (14).13See Appendix C for additional discussion of price dispersion in the model.
11
Ct = ct denotes aggregate consumption demanded by households. For simplicity, there are no
government purchases or investment in the baseline version of the model.
Finally, there is a monetary authority that sets the one-period nominal interest rate it
according to a Taylor (1993)-type policy rule,
it = r + πt + φπ(πt − π) +φy4(yt − yt), (19)
where r = − log β denotes the continuously-compounded steady-state real interest rate, πt ≡log(Pt/Pt−1) denotes the inflation rate, π the monetary authority’s inflation target, yt ≡ log Yt,
yt ≡ ρyyt−1 + (1− ρy)yt (20)
denotes a trailing moving average of log output, and φπ, φy ∈ R and ρy ∈ [ 0, 1) are parameters.14
The term (πt − π) in (19) represents the deviation of inflation from policymakers’ target and
(yt − yt) is a measure of the “output gap” in the model.15
2.4 Solution Method
I solve the model above by writing each equation in recursive form, dividing nonstationary vari-
ables (Yt, Ct, wt, etc.) by At so that the resulting ratios have a stable nonstochastic steady
state. I then use the method of local approximation around the nonstochastic steady state, or
perturbation methods, to compute a numerical solution to the model.16 For the complete set of
recursive equations that define the model and additional discussion, see Appendix A.
Macroeconomic models similar to the one developed above are typically solved using a first-
order approximation (a linearization or log-linearization), but this solution method reduces all
14Note that interest rates and inflation in (19) are at quarterly rather than annual rates, so φy corresponds tothe sensitivity of the annualized short-term interest rate to the output gap, as in Taylor (1993). I also exclude alagged interest rate “smoothing” term on the right-hand side of (19) for simplicity and to keep the number of statevariables in the model to a minimum. Rudebusch (2002) argues that the degree of federal funds rate smoothingfrom one quarter to the next is essentially zero, and that instead the Federal Reserve’s deviations from the Taylorrule (19) are serially correlated—i.e., that the residuals εit in the empirical version of (19) are serially correlated.
15This is an empirically motivated definition of the output gap: it implies that the central bank will raise short-term nominal interest rates when output rises above its recent history and lower rates when output falls below thathistory, all else equal. The behavior of monetary policy is very important for the sign and size of risk premia onnominal and real bonds in the model. In order for the model to match these risk premia in the data, it’s importantthat monetary policy act in a way that is consistent with the data. Defining the output gap to be the deviationof output from flexible-price output implies that interest rates would behave in an opposite manner, and generallywould not allow the model to match empirical risk premia on nominal and real bonds.
16The equity price pet (defined below) is normalized by Aνt rather than At, where ν denotes the degree of leverage.
The value function Vt is normalized by defining Vt ≡ Vt − logAt.
12
risk premia in the model to zero.17 A second-order approximation to the model produces risk
premia that are nonzero but constant over time (a constant function of the variance σ2A). In order
for risk premia in the model to vary with the state of the economy, the model must be solved
to at least third order around the steady state. Note that second- and third-order terms in the
model solution can be non-negligible as long as the model is sufficiently “curved”, which is the
case when risk aversion (related to the Epstein-Zin parameter α) is sufficiently large.
I compute third- and higher-order solutions of the model using the Perturbation AIM al-
gorithm of Swanson, Anderson, and Levin (2006), which can compute general nth-order Taylor
series approximate solutions to discrete-time recursive rational expectations models. The model
above has two state variables (Δt, yt) and a single shock (εAt+1) and thus can be solved to third
order very quickly, in just a few seconds on a laptop computer. To obtain greater accuracy over
a wider range of values for the state variables, the model can be solved to higher order; the re-
sults reported below are for the fifth-order solution unless stated otherwise. (Results for fourth-
and sixth-order solutions are very similar, suggesting that the Taylor series has essentially con-
verged over the relevant range for the state variables.) Aruoba et al. (2006) compare a variety
of numerical solution techniques for standard macroeconomic models and find that higher-order
perturbation solutions are among the most accurate globally as well as being the fastest to com-
pute. Swanson, Anderson, and Levin (2006) provide details of the algorithm and discuss the
global convergence properties of nth-order Taylor series approximations.
A noteworthy feature of the nonlinear solution algorithm I use here, relative to the loglinear-
lognormal approximation typically used in finance, is that second- and higher-order terms of the
Taylor series display endogenous conditional heteroskedasticity. Letting xt denote a generic state
variable and εt+1 a generic shock, the second-order Taylor series solution has terms of the form
xtεt+1, which have a one-period-ahead conditional variance that depends on the economic state xt
(that is, Vart(xtεt+1) depends on xt). Thus, even though the model’s exogenous driving shocks
εAt+1 are homoskedastic, the nonlinear solution algorithm I use here preserves the endogenous
conditional heteroskedasticity that is naturally generated by the nonlinearities in the model.
17 In the finance literature, it is standard to log-linearize the model and then take expectations of all variablesassuming joint lognormality. This approximate solution method produces nonzero (but constant) risk premia,but effectively treats higher-order moments of the lognormal distribution on par with first-order economic terms.Standard perturbation methods (e.g., Judd, 1998; Swanson, Anderson, and Levin, 2006) explicitly relate higher-order moments of the shock distribution to the corresponding order of the state variables (so variance is a second-order term, skewness a third-order term, etc.), because their magnitudes are the same in theory.
The model described above is meant to be illustrative rather than provide a comprehensive empir-
ical fit to the data, so I calibrate rather than estimate its key parameters. The baseline calibration
is reported in Table 1, and is meant to be standard, following along the lines of parameter val-
ues estimated by Smets and Wouters (2007), Altig et al. (2011), and Del Negro, Giannoni, and
Schorfheide (2015) using quarterly U.S. data.
I set the household’s discount factor, β, to .992, implying a nonstochastic steady-state real
interest rate of about 3.2 percent per year. Although this might seem a bit high, households’
risk aversion drives the unconditional mean of the risk-free real rate close to 2 percent in the
stochastic case.
The household’s logarithmic preferences over consumption imply an intertemporal elasticity
of substitution of unity, which is higher than estimates based on aggregate data (e.g., Hall, 1988),
but is similar to estimates using household-level data (e.g., Vissing-Jorgensen, 2002). Bansal and
Yaron (2004) and Dew-Becker (2012) argue that estimates based on aggregate data are biased
downward, providing further support for the value of unity used here. In addition, logarith-
mic preferences over consumption are a standard benchmark in macroeconomics (e.g., King and
Rebelo, 1999).18
The calibrated value of χ = 3 implies a Frisch elasticity of labor supply of 1/3, consistent
with estimates in Del Negro et al. (2015) and estimates from household data (e.g., MaCurdy,
1980; Altonji, 1986). I set the parameter η so as to normalize L = 1 in steady state.
The parameter α is calibrated to imply a coefficient of relative risk aversion Rc = 60
in steady state, using the closed-form expressions derived in Swanson (2018) for models with
18My results are not sensitive to setting the IES equal to unity. For example, specifications with u(ct, lt) =
c1−γt /(1−γ)−ηl1+χ/(1+χ) produce very similar results when γ is set to 0.9 or 1.1. Of course, these specificationsdo not satisfy balanced growth and are nonstationary in response to permanent technology shocks.
14
labor.19 Although this value is high, it is a well-known byproduct of the model’s simplicity:20 for
example, households in the model have perfect knowledge of all the model equations, parameter
values, shock dynamics, and shock distributions, so the quantity of risk in the model is very
small relative to the actual U.S. economy. As a result, the household’s aversion to risk in the
model must be correspondingly larger to fit the risk premia seen in the data. Barillas, Hansen,
and Sargent (2009) formalize this intuition by showing that high risk aversion in an Epstein-Zin
specification is isomorphic to a model in which households have low risk aversion but a moderate
degree of uncertainty about the economic environment.21 As an alternative to high risk aversion,
one could increase the quantity of risk in the model instead, such as by introducing long-run risk
as in Bansal and Yaron (2004), or disaster risk as in Rietz (1988) and Barro (2006).22
Turning to the production side of the economy, I set the elasticity of output with respect
to labor θ = 0.6. I calibrate the Calvo contract parameter ξ = 0.8, implying an average contract
duration of five quarters, consistent with the estimates in Altig et al. (2010) and Del Negro et
al. (2015). I calibrate the monopolistic markup λ for intermediate goods to 1.1, consistent with
the estimates in Smets and Wouters (2007) and Altig et al. (2010). The technology process At
is calibrated to be a random walk in the baseline calibration, ρA = 1. The standard deviation of
technology shocks, σA, is set to .007, as estimated by King and Rebelo (1999). The steady-state
ratio of the capital stock to annualized output is calibrated to 2.5.
The response of monetary policy to inflation, φπ, is set to 0.5, as in Taylor (1993, 1999).
I set φy = 0.75, between the values of 0.5 and 1 used by Taylor (1993) and Taylor (1999). I set
the monetary authority’s inflation target π to 0.8 percent per quarter, implying a nonstochastic
19Swanson (2018) derives the coefficient of relative risk aversion for generalized recursive preferences with flexiblelabor and arbitrary period utility function u(ct, lt). For multiplier preferences with period utility function (1) andl = 1 in steady state, risk aversion is given by Rc= α+ (1+(η/χ))−1. See Swanson (2018) for the derivation anddetails. In general, risk aversion is lower when labor supply can vary because the household is better able to insureitself from shocks.
20For example, Piazzesi and Schneider (2006) estimate a value of 57, Rudebusch and Swanson (2012) a value of110, Van Binsbergen et al. (2012), Andreasen (2012), and Campbell and Cochrane (1999) a value of about 80, andTallarini (2001) a value of about 50. The nonstationarity of technology implied by ρA = 1 in the present paperincreases the quantity of risk in the model here relative to Rudebusch and Swanson (2012), which allows me touse a lower coefficient of relative risk aversion than in their paper.
21See also Campanale, Castro, and Clementi (2010), who emphasize that the quantity of consumption riskin a standard DSGE model is very small, and thus the risk aversion required to match asset prices must becorrespondingly larger.
22The simplifying representative-household assumption could also be dropped. Mankiw and Zeldes (1991),Parker (2001), and Malloy, Moskowitz, and Vissing-Jorgensen (2009) show that the consumption of stockholdersis more volatile (and more correlated with the stock market) than the consumption of nonstockholders, so therequired level of risk aversion in a representative-agent model is higher than it would be in a model that recognizedthat stockholders have more volatile consumption (Guvenen, 2009).
15
steady-state inflation rate of about 3.2 percent per year. Although this is higher than the value
of about 2 percent used by many central banks as their current official inflation target, there
are two reasons why a higher number is appropriate here: First, a steady state inflation rate
of 2 percent is too low to explain the historical average level of nominal yields in the U.S. and
U.K. (and many other countries), even over relatively recent samples such as 1990–2018, as I will
show below. Second, households’ risk aversion drives the unconditional mean of inflation in the
stochastic version of the model somewhat below the nonstochastic steady-state value. Finally, I
calibrate ρy = 0.9, implying that the monetary authority uses the deviation of current output
from its average level over the past roughly 2.5 years to approximate the output gap.
2.6 Impulse Response Functions
Figure 1 plots impulse response functions for the model to a one-standard-deviation (0.7 percent)
positive technology shock, under the baseline calibration described above. Recall that, because
ρA = 1, the effect of the shock on productivity is permanent. The dashed red lines in each panel
report standard impulse response functions for the first-order (log-linear) solution to the model,
while the solid blue lines report impulse response functions for the nonlinear, fifth-order Taylor
series solution.23 I start by describing the linear impulse response functions (dashed red lines),
and then describe how the fifth-order impulse response functions (solid blue lines) differ from
their linear counterparts.
The top left panel of Figure 1 reports the impulse response function for consumption, Ct, to
the shock. Consumption jumps upward on impact, as higher productivity increases the supply of
output and makes households wealthier in present-value terms, increasing consumption demand.
The first-order impulse response function for Ct does not jump all the way to its new long-run
level on impact, however, because of the increase in the real interest rate (described shortly).
Instead, consumption continues to increase gradually over time to approach its new steady state.
23The impulse response functions for the fifth-order solution to the model are computed as follows: The statevariables of the model are initialized to their nonstochastic steady-state values. The impulse response function iscomputed as the period-by-period difference between a “one-shock” and a “no-shock” (baseline) scenario. In theone-shock scenario, εAt is set equal to .007 in period 1, and equal to 0 from period 2 onward. In the no-shock
scenario, εAt is set equal to 0 in every period. Agents in the model do not have perfect foresight, so they still actin a precautionary manner even though the realized shocks turn out to be deterministically equal to 0 from period1 onward. In principle, this nonlinear impulse response function can vary as one varies the initial point of thesimulation, or may scale nonlinearly with the size of the shock εAt . In practice, however, the impulse responses inFigure 2 do not vary much with the initial point and do not display much nonlinearity in the size of the shock. Forexample, the fifth-order impulse response functions to a negative 0.7 percent technology shock, which are reportedin Figure B1 in Appendix B, look very similar to the negative of the blue lines in Figure 1 (although they are abit larger in magnitude).
16
1st-order solution
5th-order solution
0 10 20 30 40 500.0
0.2
0.4
0.6
0.8
1.0percent
ConsumptionCt
10 20 30 40 50
-1.0
-0.8
-0.6
-0.4
-0.2
0.0ann. pct.
Inflation t
10 20 30 40 50
-0.5
-0.4
-0.3
-0.2
-0.1
0.0ann. pct.
Short-term nominal interest rate it
0 10 20 30 40 500.0
0.1
0.2
0.3
0.4
0.5ann. pct.
Short-term reai lnterest rate rt
10 20 30 40 50
-0.4
-0.2
0.0
0.2
0.4
percentLabor Lt
0 10 20 30 40 500.0
0.2
0.4
0.6
0.8
1.0percent
Price Dispersion t
Figure 1. First-order (dashed red lines) and fifth-order (solid blue lines) impulse response functions forconsumption Ct, inflation πt, short-term nominal interest rate it, short-term real interest rate rt, laborLt, and price dispersion Δt to a one-standard-deviation (0.7 percent) positive technology shock in themodel. See text for details.
17
The top right panel reports the impulse response for inflation, πt. The higher level of
technology reduces firms’ marginal costs of production, and monopolistic firms set their price
equal to a constant markup λ over expected future marginal costs, whenever they are able to
reset their price. Thus, inflation falls on impact (by about 0.5 percent at an annualized rate)
as those firms who are able to reset their prices do so. The response of inflation is persistent,
however, as firms’ price contracts expire only gradually.
The nominal interest rate it, in the middle left panel, is set by the monetary authority as
a function of output and inflation according to the policy rule (19). Interest rates respond more
strongly to inflation than to output, causing the nominal interest rate to decline moderately,
on net, in response to the shock, about 40 basis points (at an annual rate) on impact before
gradually returning to steady state. However, the nominal interest rate falls by less than inflation
in response to the shock, so the real interest rt rises about 5 basis points (at an annual rate) on
impact, as can be seen in the middle right panel.24 The real rate then gradually falls back to
steady state.
The response of labor, Lt, is plotted in the bottom left panel. After the technology shock,
households are wealthier in present value terms and want to consume more leisure; this tends to
push labor downward. However, because prices are sticky and firms are monopolistic, firms hire
whatever labor is necessary to satisfy output demand, which tends to push labor upward. For
the simple model here, solved to first order, the former effect dominates, causing labor to decline
slightly on net; indeed, this result is common in simple New Keynesian models, as pointed out
by Galı (1999).25
However, this is no longer true for the fifth-order solution of the model, as can be seen
by comparing the solid blue and red dashed lines in the bottom left panel. There are two main
reasons for this difference: First, price dispersion Δt increases in response to the shock—as can
be seen in the bottom right panel of Figure 1—but only for the nonlinear solution, because
the linearized version of equation (18) implies shocks have no effect on Δt.26 The increase in
price dispersion reduces the economy’s ability to produce final output efficiently, and increases
the amount of labor required to produce any given level of output (see equation 17). Indeed,
24Recall that rt is the ex ante real interest rate, so rt = it − Etπt+1 to first order.25 In more complicated, realistic models, such as Altig et al. (2011), increased demand for investment following
the technology shock is typically enough to make the increase in firms’ labor demand dominate. Alternatively,a stronger monetary policy response that drives the short-term real interest rate down in response to the shockwould cause consumption to jump above 0.7 percent on impact and lead to an increase in labor.26The linearized version of equation (18) is Δt = ξΔt−1, which implies Δt is invariant to shocks.
18
the hump shape in dispersion is clearly visible in the nonlinear impulse response function for
labor (and to a lesser extent, consumption).27 Second, the positive technology shock reduces
the volatility of households’ stochastic discount factor (SDF), for reasons discussed in detail in
Section 4, below. The lower volatility of the SDF makes households effectively less risk averse
and reduces their demand for precautionary savings, leading to an increase in consumption Ct
relative to the linear case (as can be seen in the top-left panel). Households’ greater demand
for consumption requires firms to hire more labor, putting further upward pressure on Lt in the
bottom-left panel.28
3. Asset Prices and Risk Premia
The stochastic discount factor implied by the simple macroeconomic model above can now be
used to price any asset in the model. In particular, we can derive the implications of the model
for the prices of equity and real, nominal, and defaultable debt.
3.1 Equity
I define an equity security in the model to be a levered claim on the aggregate consumption
stream. The definition of equity as a consumption claim maximizes comparability to the finance
literature and simplifies the intuition in the model; the results are very similar if equity is instead
defined to be a claim on the profits of the monopolistic intermediate firm sector.29 Each period,
equity pays a dividend equal to Cνt , where ν is the degree of leverage. Consistent with Abel
(1999) and Bansal and Yaron (2004), I calibrate ν = 3. Note that ν can be interpreted as the
sum of operational and financial leverage in the economy, where operational leverage results from
fixed costs of production for firms (Gourio, 2012).
27The effect is essentially symmetric for a negative technology shock—i.e., Δt decreases in response to a negativetechnology shock (see Figure B1 in Appendix B). In the stochastic version of the model, inflation is often belowthe nonstochastic steady state value (due to precautionary behavior by firms, discussed below), so even if thereare no shocks, Δt > 1 will hold. Negative technology shocks can thus decrease Δt.
28The higher level of consumption in the nonlinear case also causes the real interest rate to rise by more inthe middle-right panel. The higher level of labor in the nonlinear case increases firms’ marginal costs, which putsupward pressure on inflation. Inflation nevertheless falls a bit more on impact in the nonlinear case for reasonsdiscussed in Section 4, below. The response of the nominal interest rate it in the nonlinear case follows in astraightforward manner from Ct and πt, given the policy rule (19).
29This is because consumption, output, and monopolistic firm profits are very highly correlated in the model:Yt = Ct, and firms’ profits are essentially a levered claim on the output stream. Note that adding fixed costs ofproduction to the model would increase the degree of leverage.
19
Let pet denote the ex-dividend price of an equity security at time t. In equilibrium,
pet = Etmt+1(Cνt+1 + pet+1). (21)
Let Ret+1 denote the realized gross return on equity,
Ret+1 ≡ Cν
t+1 + pet+1
pet. (22)
I define the equity premium at time t, ψet , to be the expected excess return to holding equity for
one period:
ψet ≡ EtR
et+1 − ert . (23)
Note that
ψet =
Etmt+1Et(Cνt+1 + pet+1)−Etmt+1(C
νt+1 + pet+1)
petEtmt+1
=−Covt(mt+1, R
et+1)
Etmt+1
= −Covt
( mt+1
Etmt+1, Re
t+1
), (24)
where Covt denotes the covariance conditional on information at time t.30
The recursive equity pricing and equity premium equations (21)–(23) can be appended to
the equations of the macroeconomic model in the previous section, allowing the equity premium
(23) to be solved numerically along with the rest of the model. For the baseline calibration of the
model in Table 1, solved to fifth order, the expected excess return on equity is about 1.05 percent
per quarter (4.19 percent per year), evaluating the model’s state variables at their nonstochastic
steady-state values. Empirical estimates of the equity premium typically range from about 3
to 6.5 percent per year (e.g., Campbell, 1999, Fama and French, 2002), so the equity premium
implied by the model is consistent with the data.
The model-implied equity premium is very sensitive to both the level of risk aversion Rc
and the persistence of the technology shock ρA. Table 2 reports values for the equity premium
for several different values of Rc and ρA, holding the other parameters of the model fixed at their
baseline values in Table 1. The equity premium increases about linearly along with the household’s
30 If mt+1 and Ret+1 are jointly lognormally-distributed, as is typically assumed in the finance literature, then
the equation Etmt+1Ret+1 = 1 implies Etret+1− rft = −Covt(logmt+1, r
et+1)− 1
2Vartret+1, where ret+1≡ logRe
t+1.
Equation (24) says essentially the same thing without assuming joint lognormality.
20
Table 2: Equity Premium as a Function of Risk Aversion and Shock Persistence
Model-implied equity premium ψe, in annualized percentage points, for different values of relative riskaversion Rc and technology shock persistence ρA, holding the other parameters of the model fixed attheir baseline values from Table 1. State variables of the model are evaluated at the nonstochastic steadystate. See text for details.
coefficient of relative risk aversion, Rc, consistent with the analysis in Swanson (2018).31 Perhaps
more surprising is the substantial drop in the equity premium for values of ρA that are only
slightly less than unity—for example, reducing ρA from 1 to .995 reduces the equity premium by
more than half, and reducing ρA from .995 to .99 cuts the equity premium almost in half again.
There are two reasons why ψe is so sensitive to ρA: First, equity is very long-lived, so it is sensitive
to changes in the consumption dividend even at distant horizons. Second, the household’s value
function Vt, which enters into the stochastic discount factor (6), is also sensitive to consumption
at long horizons. Reductions in ρA below unity have a very large effect on consumption at distant
horizons, and thus significantly attenuate the response of both the equity price and the SDF to
a technology shock. The subsantially lower covariance between these two variables reduces the
equity premium (equation (24)). (Note that the long-run risks literature, beginning with Bansal
and Yaron, 2004, assumes that long-run consumption is even more volatile than my baseline
random walk calibration of ρA = 1; as a result, their models imply a larger equity premium than
my model here, or a similar-sized equity premium with a lower degree of risk aversion.)
The equity premium in the model also varies substantially over time. Figure 2 plots the
nonlinear (fifth-order) impulse response functions for the equity price pet and the equity premium
ψet to the technology shock, computed the same way as the nonlinear impulse response functions
in Figure 1. The left-hand panel of Figure 2 depicts the response of the equity price, which jumps
31The equity premium increases linearly with risk aversion to second order around the nonstochastic steadystate. The equity premium in Table 2 is computed to fifth order and thus is not strictly linear in risk aversion,but the intuition from the analysis in Swanson (2018) still holds.
21
0 10 20 30 40 500.0
0.5
1.0
1.5
2.0
2.5
percentEquity price pte
10 20 30 40 50
-80
-60
-40
-20
0ann. bp
Equity premium t
e
Figure 2. Nonlinear impulse response functions for the equity price pet and equity premium ψet to a one-
standard-deviation (0.7 percent) positive technology shock in the model, with state variables initializedto their nonstochastic steady state values. See text for details.
about 2.5 percent in response to the technology shock on impact. The risk-neutral increase in
the equity price would be about 2.1 percent (the leverage ratio times the increase in dividends
of about 0.7 percent every period); the additional 0.4 percent increase in the price is due to the
decline in the risk premium that investors require to hold the risky asset. This can be seen in the
right-hand panel of Figure 2, where the equity premium drops about 60 basis points (bp) at an
annual rate on impact before rising slowly back toward its initial level. Thus, the model produces
an equity premium that is countercyclical, consistent with conventional wisdom in the literature
(e.g., Fama and French, 1989; Campbell and Cochrane, 1999; Cooper and Priestley, 2008). The
reason for this countercyclicality is that the volatility of the households’ stochastic discount factor
falls after a positive technology shock (and increases after a negative shock), for reasons I discuss
shortly below. Over the course of a year, the standard deviation of the equity premium in the
model is about 103 bp, obtained by summing the squares of the first four quarters of the impulse
response and taking the square root. (Note that this is the standard deviation of the expected
excess return on equity; I discuss the standard deviation of the ex post excess return shortly.)
To compare the model-implied time variation in the equity premium to the data, it is
useful to compute the model-implied Sharpe ratio, ψet
/√Vartret+1 . The average quarterly (non-
annualized) Sharpe ratio in the model is 1.05/2.5 = 0.42, which is about in line with the typical
estimates of 0.2 to 0.4 in the literature (e.g., Campbell and Cochrane, 1999; Lettau and Ludvigson,
2010). The fact that the model’s Sharpe ratio is at the high end of this range is not surprising
since the model is driven by a single shock and thus understates the overall volatility of equity
22
prices; adding a monetary policy shock to the model, for example, would increase the volatility
of equity without much altering its excess return (because monetary policy shocks are much less
persistent than technology shocks and have only a small effect on the equity premium), and lead
to a lower Sharpe ratio more in line with the data.
The quarterly standard deviation of the (non-annualized) Sharpe ratio in the model is about
0.62/2.5 = 0.25. Again, this is in line with estimates of the quarterly standard deviation of the
Sharpe ratio in the literature, which range between 0.09 and 0.47 (e.g., Campbell and Cochrane,
1999; Lettau and Ludvigson, 2010, Table 11.7).
The ex post excess return on equity in the model has a quarterly standard deviation of about
2.5 percent per quarter, or 5 percent per year. Empirical estimates in the literature are typically
in the range of 6 to 12 percent per quarter, or 12 to 24 percent per year (e.g., Campbell, 1999;
Lettau and Ludvigson, 2010), so the model-implied volatility for equity returns is substantially
lower than the data. However, this is again not surprising, given that the model is driven by a
single shock. Adding additional shocks to the model, such as fiscal or monetary policy shocks as
in the New Keynesian DSGE literature (e.g., Smets and Wouters, 2007), would bring equity price
volatility closer to the data.
From equation (24), we know that the decline in the model-implied equity premium in Fig-
ure 2 must be due to a drop in the conditional covariance of the equity price with the stochastic
discount factor. In other words, the model generates endogenous conditional heteroskedasticity
in response to shocks, even though the exogenous technology shock that drives the model is ho-
moskedastic. This is a striking and very important feature of the model. I discuss how the model
generates heteroskedasticity in detail in Section 4, below (and see also Section 2.4, above), but
the key factor is the behavior of price dispersion Δt. In response to a technology shock, price
dispersion moves in the same direction as the shock (see the bottom-right panels of Figures 1
and B1). Because greater price dispersion reduces output and aggregate productivity (see equa-
tion 17), the response of price dispersion to a technology shock tends to dampen the responses
of the model’s other variables to the shock. In addition, the sensitivity of price dispersion to a
technology shock is greater if there has been a positive technology shock in the recent past—see
Section 4, below. Thus, a positive technology shock today leads to a greater sensitivity of price
dispersion Δt to future shocks, which reduces the volatility of the other variables of the model
(such as consumption and the SDF) to future shocks.
The result is that the stochastic discount factor displays substantial endogenous conditional
23
heteroskedasticity. In a perfectly homogeneous, homoskedastic model—such as the ones typically
used in finance that have no labor and no nominal rigidities—there is no endogenous conditional
heteroskedasticity. The only way to generate a time-varying equity premium in those models is
to assume that the exogenous driving shock itself is conditionally heteroskedastic (e.g., Bansal
and Yaron, 2004).
3.2 Real and Nominal Default-Free Bonds
A default-free zero-coupon real bond in the model pays one unit of consumption at maturity. Let
p(n)t denote the nominal price of an n-period zero-coupon real bond, and p
(n)t ≡ p
(n)t /Pt its real
price, with p(0)t ≡ 1. Then for n ≥ 1,
p(n)t = Etmt+1p
(n−1)t+1 (25)
in equilibrium in each period t. In particular, p(1)t = e−rt .
A default-free zero-coupon nominal bond pays one nominal dollar at maturity. Let p$(n)t de-
note the nominal price of an n-period zero-coupon nominal bond, with p$(0)t ≡ 1. Then for n ≥ 1,
p$(n)t = Etmt+1e
−πt+1p$(n−1)t+1 (26)
in each period t. In particular, p$(1)t = e−it .
Let r(n)t denote the n-period continuously-compounded yield to maturity on a real zero-
coupon bond, and i(n)t the corresponding yield on an n-period nominal bond. Then
r(n)t = − 1
nlog p
(n)t (27)
and
i(n)t = − 1
nlog p
$(n)t . (28)
Note that even though these bonds are free from default, they are risky in the sense that their
prices can fluctuate in response to shocks, for n > 1.
The risk premium on a bond is typically written as a term premium, the difference between
the yield to maturity on the bond and the hypothetical, risk-neutral yield to maturity on the
same bond. The risk-neutral real price p(n)t of an n-period zero-coupon real bond is given by
p(n)t = e−rtEt p
(n−1)t+1 , (29)
24
where p(0)t ≡ 1. The n-period real term premium ψ
(n)t is then
ψ(n)t ≡ 1
n
(log p
(n)t − log p
(n)t
)(30)
The formula for the term premium on a nominal n-period bond, ψ$(n)t , is analogous.
Note that, to first order, ψ(n)t ≈ (p
(n)t − p
(n)t )/ (np(n)) , where p(n) denotes the steady-state
real bond price, and
p(n)t − p
(n)t = Etmt+1Etp
(n−1)t+1 − Etmt+1p
(n−1)t+1
= −Covt(mt+1, p
(n−1)t+1
)+ e−rtEt
(p(n−1)t+1 − p
(n−1)t+1
)= −Et
n−1∑j=0
e−rt,t+jCovt+j
(mt+j+1, p
(n−j−1)t+j+1
), (31)
where rt,t+j ≡∑t+j
τ=t+1 rτ and the last line of (31) follows from forward recursion. Equation (31)
shows that, even though the bond price depends only on the one-period-ahead covariance between
the stochastic discount factor and next period’s bond price, the risk premium on the bond depends
on this covariance over the entire lifetime of the bond. The term premium ψ(n)t then satisfies
ψ(n)t ≈ − 1
np(n)Et
n−1∑j=0
e−rt,t+jCovt+j
(mt+j+1, p
(n−j−1)t+j+1
). (32)
Intuitively, the term premium is larger the more negative the covariance between the SDF and
the price of the bond over the lifetime of the bond.
The bond pricing and bond yield equations (25)–(30) are recursive and can be appended to
the macroeconomic model described above and solved numerically along with the macroeconomic
variables, equity price, and equity premium. (Note that, to consider a bond with n periods to
maturity, n − 1 bond pricing equations must be appended to the model, one for each maturity
from 2 to n.)
Table 3 reports the real yield curve implied by the model, along with the corresponding
average real yields estimated from inflation-indexed government bonds in the U.S. and U.K. over
different sample periods. Data for U.S. inflation-indexed Treasuries (TIPS) are taken from the
updated Gurkaynak, Sack, and Wright (2010) online dataset. The first TIPS were issued in 1998,
and a yield curve for maturities of 5 years or more can be estimated beginning in 1999. The
first row of Table 3 thus reports average TIPS yields from 1999 to 2018. Real yields over this
sample averaged about 1.15 to 1.65 percent per year. Zero-coupon yields for shorter-maturity
25
Table 3: Real Zero-Coupon Bond Yields, Data vs. Model
aGurkaynak, Sack, and Wright (2010) online dataset.bEvans (1999).cBank of England web site.
Estimated zero-coupon real yields from inflation-indexed bonds in the U.S. and U.K., and zero-couponreal yields implied by the macroeconomic model presented above. The last column reports the differencebetween the 10-year and 3-year yields in each row. See text for details.
TIPS (down to 2 years; neither Gurkaynak et al., 2010, nor the Bank of England report zero-
coupon real yields with a maturity less than 2 years) can be estimated beginning in 2004, and are
reported in the second row of Table 3, along with the average yields for longer maturities over
the same sample. This sample also excludes the period of lower TIPS liquidity in the first few
years after they were introduced. Over this sample, average real yields are lower, between about
0.1 and 1.1 percent. However, the period from 2008–15 is unusual in that the financial crisis and
severe recession led the Federal Reserve to reduce short-term interest rates to record lows, and to
some extent we might expect this to show up in shorter-term real yields as well, both as a lower
level of yields and as a steeper yield curve slope. Thus, the third row of Table 3 reports results
from 2004–07, a short sample, but one that avoids both the low liquidity of TIPS in its first few
years and the financial crisis and recession. Over this sample, real yields average between about
1.4 and 2.1 percent.
However, this is a short sample and the period from 2004–05 was also characterized by
easy monetary policy and a low level of short-term U.S. yields as the Federal Reserve worked to
facilitate recovery from the 2001 recession. Thus, the next three rows of Table 3 report average
real yields on inflation-indexed gilts in the U.K., for which we have a longer sample. Evans (1999)
estimates real zero-coupon U.K. yields from 1983 to 1995, reported in the fourth row of Table 3,
which average between about 4 and 6 percent over that sample. Interestingly, the real U.K. gilt
yield curve slopes downward rather than upward over this period, by about 100–200 basis points.
However, as in the U.S., the early years of the U.K. indexed gilt market may have suffered from
26
low liquidity and correspondingly higher yields. Thus, the fifth row of Table 3 reports estimated
real yields from 1985 to 2018, from the Bank of England’s web site. Over this longer sample, real
U.K. yields average about 1.5 to 1.9 percent, and the yield curve sloped upward by about 37 bp.
The sixth row of Table 3 reports results for the U.K. excluding both the early years of the sample
and the financial crisis and recession period. Over this sample, 1990–2007, real yields in the U.K.
are a bit higher, averaging about 2.8 percent, and the yield curve is about flat, sloping upward
by 1 bp.
While the exact level and slope of the real yield curve depend on the sample period and
country considered (U.S. vs. U.K.), the overall pattern suggests an average real yield of approx-
imately 2 percent per year, with a slope that is relatively flat—neither strongly upward-sloping
nor downward-sloping on average. The macroeconomic model developed in this paper fits these
features of real yields in the data quite well. Real yields in the model average a bit less than 2
percent under the baseline calibration, evaluating the model’s state variables at the nonstochastic
steady state.32 The model also implies that the real yield curve is about flat on average, with
essentially no spread between the 10-year and 3-year yields, and a −1 bp spread between the
10-year and 2-year yields.
A downward-sloping real yield curve is a standard feature of traditional real-business-cycle
models—see Backus, Gregory, and Zin (1989), Donaldson, Johnsen, and Mehra (1990) and Den
Haan (1995). Intuitively, if short-term real interest rates fall in recessions, then the price of a
long-term real bond will tend to rise in recessions, which is when households value consumption
the most. Thus, long-term real bonds act like recession insurance and should carry a negative
risk premium. In the macroeconomic model I develop here, the response of the short-term real
interest rate to the shock is fairly small (Figure 1), implying a relatively small change in the real
long-term bond price (Figure 3, below). Moreover, the fall in the real term premium after the
shock (Figure 3) attenuates the response of the real bond price even more. As a result, the price
of a real long-term bond is not very countercyclical and the insurance properties of the bond are
minor, resulting in only a small risk premium, consistent with the data.
Figure 3 reports nonlinear impulse response functions for the 10-year real bond price and
term premium, computed in the same way as in Figures 1 and 2. The bond price falls only about
32Note that the baseline value of β from Table 1 implies a real yield of a little more than 3 percent in thenonstochastic steady state. However, the real yield rt = 1/Etmt+1, and Etmt+1 is substantially greater than1/β in the stochastic case due to Jensen’s inequality terms. Intutively, households’ aversion to risk drives uptheir demand for precautionary savings in the riskless asset, lowering the risk-free rate below its nonstochasticsteady-state value.
27
10 20 30 40 50
-1.0
-0.8
-0.6
-0.4
-0.2
0.0percent
Real long-term bond price pt (40)10 20 30 40 50
-10
-8
-6
-4
-2
0ann. bp
Real term premium t
(40)
Figure 3. Nonlinear impulse response functions for real long-term bond price p(40)t and real term premium
ψ(40)t to a one-standard-deviation (0.7 percent) positive technology shock in the model, with state variables
initialized to their nonstochastic steady state values. See text for details.
0.2 percent on impact, due to the small increase in short-term real rates in Figure 1, and the
offsetting fall in the real term premium (right-hand panel of Figure 3). The real term premium
declines in response to the shock, but by much less than the equity premium, only about 4.5 bp.
Table 4 compares the nominal yield curves implied by the model to the data. Gurkaynak,
Sack, and Wright (2007) estimate zero-coupon nominal Treasury yields for the U.S. going back
to 1961 for maturities out to 7 years, and 1971 for maturities out to 10 years (data through the
present are available from the online version of their dataset). From 1961 to 2018, nominal yields
averaged about 5.1 to 6 percent. From 1971 to 2018, the average is a bit higher, about 5.15 to 6.45
percent, with an average yield curve slope of about 130 bp. Just as for real yields, though, the
period from 2008–18 may be atypical in that short-term interest rates hit record lows in response
to the financial crisis and recession, and were constrained by the zero lower bound on nominal
rates. The “Great Inflation” period of the 1970s and early 1980s may also be problematic in that
monetary policy may have experienced a structural break since that period and is now conducted
in a more aggressive anti-inflationary manner (e.g., Clarida, Galı, and Gertler, 1999). Thus, the
third row of Table 4 reports average yields from 1990 to 2007, a period that excludes both the
Great Inflation and recent Great Recession periods. Over this sample, nominal Treasury yields
averaged about 4.5 to 6 percent, with a yield curve slope of about 140 bp.
The Bank of England also reports estimated zero-coupon yield curves for the U.K. going
back to 1970. From 1970 to 2018, nominal gilt yields in the U.K. averaged between about 6.5
and 7.5 percent, with a yield curve slope of about 100 bp, as reported in Table 4. Restricting
28
Table 4: Nominal Zero-Coupon Bond Yields, Data vs. Model
1-yr. 2-yr. 3-yr. 5-yr. 7-yr. 10-yr. (10y)−(1y)
US Treasuries, 1961–2018a 5.07 5.29 5.48 5.76 5.97
UK gilts, 1990–2007b 6.20 6.29 6.38 6.47 6.50 6.48 0.28
macroeconomic model 5.35 5.59 5.80 6.09 6.27 6.44 1.09
aGurkaynak, Sack, and Wright (2007) online dataset.bBank of England web site.
Empirical estimates of zero-coupon nominal yields from government bonds in the U.S. and U.K., andzero-coupon nominal yields implied by the macroeconomic model presented above. The last columnreports the difference between the 10-year and 1-year yield in each row. See text for details.
attention to the period from 1990 to 2007, average U.K. nominal yields are a bit lower, about 6.2
to 6.5 percent, with a slope of just 28 bp.
Again, the exact level and slope of the nominal yield curve depends on the sample period
and country considered, but nominal yields appear to average about 5 or 6 percent and have
an upward slope of about 100 bp. The model is able to reproduce these features of the data
quite well: the average level of nominal yields in the model (evaluating the state variables at the
nonstochastic steady state) is between about 5.4 and 6.4 percent, with an upward slope of 109 bp.
Thus, although the model-implied real yield curve is flat, the implied nominal yield curve slopes
upward substantially. As discussed by Rudebusch and Swanson (2012), this is because technology
shocks in the model make nominal bonds risky: A negative technology shock causes inflation to
rise persistently at the same time that consumption falls; as a result, long-term nominal bonds
in the model lose value in recessions. This implies that long-term nominal bonds should carry
a substantial risk premium, about 109 bp over the corresponding risk-neutral yield. Thus, the
simple model presented here provides a straightforward answer to the puzzle posed by Backus,
Gregory, and Zin (1989), Donaldson, Johnsen, and Mehra (1990), and Den Haan (1995): namely,
why does the nominal yield curve slope upward? The answer is technology shocks, or more
generally, any “supply shock” that causes inflation and output to move in opposite directions,
such as an oil price shock or markup shock.
Of course, the larger and more important are technology or supply shocks in the model,
the larger the term premium on nominal bonds. Thus, if supply shocks were relatively larger in
the 1970s and early 1980s than in the 1960s or more recently, the model predicts that we should
29
0 10 20 30 40 500.0
0.5
1.0
1.5
2.0percent
Nominal long-term bond price pt$(40)
10 20 30 40 50
-10
-8
-6
-4
-2
0ann. bp
Nominal term premium t
$(40)
Figure 4. Nonlinear impulse response functions for nominal long-term bond price p$(40)t and term
premium ψ$(40)t to a one-standard-deviation (0.7 percent) technology shock in the model, with state
variables initialized to their nonstochastic steady state values. See text for details.
see a larger term premium on nominal bonds in those periods when supply shocks were larger.
In fact, this prediction seems to be consistent with the data: Rudebusch, Sack, and Swanson
(2007) graph several measures of the term premium—from a VAR, affine no-arbitrage models
with latent or observable factors, and the Cochrane-Piazzesi (2005) “tent-shaped” predictor of
excess returns—and for all of these measures, the estimated term premium on long-term nominal
bonds in the U.S. is higher in the 1970s and early 1980s than in the 1960s or more recently.
Campbell, Sundaram, and Viceira (2013) also document changes in the correlation between
stock and nominal bond returns over time. Although the baseline calibration of the model here has
only a single shock, making it stochastically singular, extending the model to include fiscal and/or
monetary policy shocks is straightforward and is standard in the medium-scale New Keynesian
DSGE literature (e.g., Smets and Wouters, 2007). In these models, if the relative importance
of technology or supply shocks is varied, then the size of the term premium and the correlation
between stock and bond returns will vary as well. Thus, changing correlations between stock and
bond returns can be mapped back to more structural features of the model.
Figure 4 plots the nonlinear impulse response functions for the 10-year nominal bond price
and term premium to a one-standard-deviation positive technology shock, computed in the same
way as in Figures 1–3. As discussed above, a positive technology shock causes inflation and
the short-term nominal interest rate to fall (Figure 1) and the nominal long-term bond price
to rise substantially (Figure 4), about 1.7 percent on impact before gradually returning back to
steady state. The nominal term premium falls about 10 bp on impact, so part of the strong
30
price response of the long-term nominal bond is due to the fall in the term premium. The reason
for that fall is essentially the same as for the equity premium: the decline in the volatility of
the households’ stochastic discount factor after the positive technology shock. Importantly, the
model’s prediction of a countercyclical term premium is consistent with the evidence in Fama
and French (1989), Cooper and Priestley (2008), Piazzesi and Swanson (2008), and conventional
wisdom in the literature (e.g., Campbell and Cochrane, 1999). Over the course of a year, the
standard deviation of the term premium is about 16 bp.
Estimates of the quarterly standard deviation of the term premium in the data range be-
tween about 8 to 40 bp: standard affine term structure models with latent factors, such as Kim
and Wright (2005), imply a quarterly standard deviation of about 30–35 bp, but Rudebusch
and Wu (2007) argue that these highly-parameterized models tend to overfit the high-frequency
fluctuations in long-term yields, and that fluctuations in the term premium are smaller, only
about 8 bp from quarter to quarter (see also the survey of empirical estimates in Rudebusch,
Sack, and Swanson, 2007). The term premium implied by the model of the present paper is
consistent with this range of estimates, but lies toward the lower end, consistent with the less
highly-parameterized models of Rudebusch and Wu (2007) and others.
3.3 Defaultable Bonds
In the interest of simplicity, I model a defaultable bond as a depreciating nominal consol that has
some probability of defaulting each period, an approach that has also been used in the finance
literature (e.g., Leland, 1994, 1998; Duffie and Lando, 2001; Chen, 2010). The credit spread in the
model is the difference in yield between the defaultable consol and an otherwise identical consol
that is free from default. I consider two cases in the analysis below: first, where the probability
of default and the recovery rate given default are constant over time, and second, where those
quantities vary cyclically in line with the data.
A default-free depreciating nominal consol is an infinitely-lived bond that pays a geometri-
cally declining coupon of δn−1 dollars in each period n = 1, 2, . . . after issuance. The equilibrium
ex-coupon price pct of the consol in period t is given by
pct = Etmt+1e−πt+1(1 + δpct+1), (33)
where the size of the next coupon payment is normalized to one dollar. The very simple recursive
structure of (33) makes this type of long-term bond extremely convenient to work with in a
31
model.33 When δ = 0, the consol reduces to a one-period zero-coupon bond, and when δ = 1,
it behaves like a traditional nondepreciating consol. By choosing δ appropriately, the consol can
be given any desired Macauley duration and made to behave very similarly to the corresponding
zero-coupon bond.
The continuously-compounded yield to maturity, ict , for the consol satisfies
pct =1
eict+
δ
e2ict+
δ2
e3ict+ · · · , (34)
implying
ict = log( 1
pct+ δ
). (35)
The Macauley duration of the consol is
−d log pct
dict= 1 + δpct . (36)
When calibrating the model below, I set δ so that the consol has a Macauley duration of 10 years,
corresponding to the approximate duration of the longer-term coupon bonds in Moody’s indexes.
A defaultable consol pays a nominal coupon each period in the same way as a default-free
consol, but in addition there is a chance each period that the bond will default and cease paying
interest forever. In the event of default, bondholders receive a recovery rate times the previous
value of the bond, which we can calibrate to the data. Thus, the defaultable consol price pdt
satisfies
pdt = Etmt+1e−πt+1
[(1− 1d
t+1)(1 + δpdt+1) + 1dt+1 ωt+1 p
dt
], (37)
where 1dt is an indicator variable equal to 1 if the bond defaults in period t and 0 otherwise, and
ωt denotes the recovery rate on the bond in the event of default. The yield to maturity idt of the
defaultable bond is defined by equation (35), with pdt in place of pct , and the credit spread is the
yield differential, idt − ict .
It remains to calibrate Prt{1dt+1 = 1} and ωt in (37). The average rate of default for bonds
initially rated Baa or BBB is about 0.6 percent per year (Moody’s, 2006; Standard & Poor’s, 2014),
and the average recovery rate on defaulted bonds is about 42 percent (Chen, Collin-Dufresne, and
33Leland (1994), Duffie and Lando (2001), and Chen (2010) use a nondepreciating consol to model corporatebonds, while Leland (1998) uses a depreciating consol. Rudebusch and Swanson (2008) use a (default-free) depre-ciating consol to study the long-term bond premium puzzle. The behavior of the depreciating consol in the simplemodel above and in Rudebusch and Swanson (2008) is very similar to that of a zero-coupon bond with the sameMacauley duration.
32
Table 5: Model-Implied Credit Spread on Defaultable Bonds
average ann. cyclicality of average cyclicality of creditdefault prob. default prob. recovery rate recovery rate spread (bp)
.006 0 .42 0 34.0
.006 −0.3 .42 0 130.9
.006 −0.3 .42 2.5 143.1
.006 −0.15 .42 2.5 78.9
.006 −0.6 .42 2.5 367.4
.006 −0.3 .42 1.25 137.0
.006 −0.3 .42 5 155.2
Model-implied credit spread idt − ict for defaultable vs. default-free depreciating consols with Macauleyduration of 10 years. Average annualized default probability is calibrated to bonds initially rated Baa.Cyclicality of default probability and recovery rate are the loadings on the output gap, yt − yt. See textfor details.
Goldstein, 2009; Chen, 2010).34 As a first calibration, I set Prt{1dt+1 = 1} to a constant rate of
0.15 percent per quarter and ωt to a constant of 42 percent.
The credit spread implied by the model for this calibration is reported in the first row of
Table 5. With a constant average annual default probability of 0.6 percent, the model-implied
credit spread is about 34 bp. This is essentially the risk-neutral expected loss each period from
default, (.006)(.58) = 34.8 bp, and is far less than the historical average credit spread on Baa-
rated bonds of about 120 bp (e.g., Chen, Collin-Dufresne, and Goldstein, 2009; Chen, 2010).35
Intuitively, if the risk of default in the model is uncorrelated with the stochastic discount factor,
there is no additional risk premium attached to losses from default.
Empirically, however, corporate bond defaults are highly countercyclical and recovery rates
highly procyclical (e.g., Chen, 2010; Giesecke, Longstaff, Schaefer, and Strebulaev, 2011; Standard
& Poor’s, 2011). For example, as reported by Chen (2010), the default rate averages about 0.9
percent for all bonds over the postwar period, but spikes to about 3.7 percent in 1990, 4 percent
in 2001, and 5.5 percent in 2009, with smaller increases in earlier recessions (and a spike to 8.5
percent in 1933). In boom years, the default rate falls to essentially zero. Recovery rates average
34The default rate on bonds currently rated Baa/BBB is much lower, about 0.15 percent per year on average.However, these bonds also lose value when they are downgraded, which happens with much higher probabilitythan default. Rather than keep track of credit ratings, the probability of downgrades, and capital losses in theevent of downgrade, I simply keep track of the default rate for bonds initially rated Baa/BBB.35This is the average difference between the yield on Moody’s Baa and Aaa seasoned corporate bond indexes
from 1921–2013. The average spread over alternative sample periods is similar. The spread between Baa-ratedcorporate bonds and U.S. Treasuries is even larger, about 185 bp. However, U.S. Treasuries carry an additionalpremium for their extreme liquidity and beneficial tax treatment, so the Baa-Aaa spread is often used instead,since Aaa corporate bonds are similar in liquidity and tax treatment to Baa-rated bonds and the probability ofdefault on Aaa-rated bonds is still extremely low (e.g., Chen et al., 2009).
33
about 42 percent, but drop to about 20–25 percent in 1990, 2001, and 2009, and rise to 50–60
percent in boom years.
Thus, the next rows of Table 5 consider cases where the default rate, recovery rate, or
both are correlated with the output gap in the model, yt − yt. I calibrate the cyclicality of the
model’s annualized default rate to −0.3, which implies a drop in output of 5 percent below trend
is associated with an increase in the default rate of about 1.5 percentage points.36 While this
cyclicality is lower than in Chen (2010), my focus here is on bonds initially rated Baa/BBB,
while the data in Chen (2010) is for all bonds, which includes many that were issued at ratings
below investment grade. The second row of Table 5 reports the model-implied credit spread when
the default rate is countercyclical, holding the recovery rate constant over time. This greatly
increases the model-implied credit spread, to about 131 bp, consistent with the observed spread
in the data.
The third row of Table 5 considers the case where the recovery rate is also cyclical. I
calibrate the cyclicality of the recovery rate in the model to 2.5, so that a fall in output of 5
percent below trend is associated with a roughly 12.5-percentage-point decrease in the recovery
rate on defaulted corporate bonds, in line with the variation reported in Chen (2010). Given this
degree of cyclicality, the credit spread in the model increases a bit further, to 143 bp, still close
to (and even a bit above) the value of 120 bp in the data.
In the last four rows of Table 5, I vary these cyclicality parameters to check how sensitive
the credit spread is to their variation. Changes in the cyclicality of default have a large effect
on the spread, while changes in the cyclicality of the recovery rate have a much smaller effect,
moving the spread by only a few basis points. Intuitively, a marginal increase in the probability of
default is much more costly to households because it implies an increase in the chance of a large
loss; in contrast, a marginal fall in the recovery rate implies only a small chance (0.15 percent
per quarter) of a modest increase in the loss. Thus, the cyclicality of recovery rates is much less
important in the model and can largely be ignored.
Figure 5 reports nonlinear impulse response functions for the defaultable bond price and
credit spread to a positive one-standard-deviation technology shock, computed the same way
as in previous figures. The default probability and recovery rate in the model are assumed to
have the same cyclicality as in the third row of Table 5, consistent with the data. On impact,
36To prevent the default rate in the model from becoming negative, I model it in logarithms rather than inlevels. That is, the cyclicality of the log default rate is set to −50, which, when multiplied by the average defaultrate of .006 per year, produces −0.3.
34
0 10 20 30 40 500.0
0.5
1.0
1.5
2.0
2.5percent
Nominal defaultable long-term bond price ptd
10 20 30 40 50
-10
-8
-6
-4
-2
0ann. bp
Credit spread itd - itc
Figure 5. Nonlinear impulse response functions for defaultable long-term bond price pdt and credit spreadidt − ict to a one-standard-deviation (0.7 percent) technology shock in the model, with state variablesinitialized to their nonstochastic steady state values. See text for details.
the defaultable bond price jumps about 2.3 percent, in between the responses of the default-free
nominal bond and equity prices in Figures 4 and 2. This is intuitive, since defaultable bonds are
riskier than default-free bonds but less risky than equity in the model.
The credit spread, depicted in the right-hand panel of Figure 5, drops about 7 bp on impact.
This is somewhat less than in the data; the standard deviation of the post-war quarterly change
in the Baa-Aaa spread is about 20 bp. However, as discussed above, the model here has only
one driving shock; extending the model to include additional shocks would increase the overall
volatility of the credit spread and bring it closer to the data.
To some extent, the model’s ability to jointly fit equity returns and corporate bond yields
is not surprising, since Chen, Collin-Dufresne, and Goldstein (2009), Bhamra, Kuehn, and Stre-
bulaev (2010), and Chen (2010) achieve similar simultaneous fits in an endowment economy.
Nevertheless, the present paper is the first to jointly match these data in a fully-specified macroe-
conomic model. The distinction is important because results for asset prices in an endowment
economy do not necessarily carry over to the case where households can vary their consumption
(and SDF) endogenously in response to shocks, as discussed in the Introduction. Like the present
paper, Bhamra et al. (2010) and Chen (2010) use Epstein-Zin preferences, albeit with consump-
tion and inflation taken to be exogenous, reduced-form processes. The advantage of the structural
macroeconomic approach I take here is its greater robustness to structural breaks and ability to
consider novel policy interventions, which cannot be studied in a reduced-form macroeconomic
environment. On the other hand, the much simpler macroeconomic structure in Bhamra et al.
35
(2010) and Chen (2010) allows them to perform a more structural analysis of firms’ corporate fi-
nancing and default decisions. In other words, I have adopted a simplistic, reduced-form model of
the firm in order to better focus on the structural behavior of the macroeconomy, while Bhamra et
al. (2010) and Chen (2010) have adopted a simplistic, reduced-form model of the macroeconomy
to better focus on the structural finance behavior of the firm.
4. Discussion and Extensions
The macroeconomic model developed above is essentially a “proof of concept” that a standard
New Keynesian model with Epstein-Zin preferences is consistent with a wide variety of asset
pricing facts. I now discuss several features of the model in greater detail. First, I examine
how the model produces endogenous conditional heteroskedasticity in response to homoskedastic
exogenous shocks. Second, I discuss the relationship between the conditional heteroskedasticity
in the model and the literature on “uncertainty shocks” (e.g., Bloom, 2009); I also compare the
model’s unitary intertemporal elasticity of substitution to the typical assumption that the IES
> 1 in the long-run risks literature. Third, I extend the model to include additional shocks, such
as a monetary policy shock and a fiscal policy shock, and discuss how this affects the results.
Finally, I discuss the implications of the model’s ability to price assets endogenously for financial
frictions models, such as Bernanke, Gertler, and Gilchrist (1999), Kiyotaki and Moore (1997),
and others.
4.1 Endogenous Conditional Heteroskedasticity
The macroeconomic model developed above displays endogenous conditional heteroskedasticity in
response to shocks, as mentioned in Sections 2.4 and 3.1, above. Indeed, this feature of the model
is crucial for generating time-varying risk premia: If the stochastic discount factor and an asset
return are both homoskedastic, then the risk premium on the asset must be constant over time,
as could be seen in equation (24). The fact that the model produces time-varying risk premia in
Figures 2 through 5 is therefore evidence that the model-implied SDF is heteroskedastic.
In Figure 6, I analyze this heteroskedasticity in more detail. The SDF in the model
is given by mt+1 = β(Ct+1/Ct)−1
[exp(−αVt+1)/Et exp(−αVt+1)
], so its conditional volatil-
ity, Vart logmt+1, can be decomposed into two parts: the variance of consumption growth,
Vart log(Ct+1/Ct)−1, and the variance of the Epstein-Zin component, Vart log
[exp(−αVt+1)/
36
10 20 30 40 50
-50
-40
-30
-20
-10
0percent
Conditional Variance Vart [(Ct+1/Ct)-1]10 20 30 40 50
-50
-40
-30
-20
-10
0percent
Conditional Variance Vart [exp(- Vt+1/)Etexp(- Vt+1/]
Figure 6. Nonlinear impulse response functions for conditional variances Vart(Ct+1/Ct)−1 and
Vart[ exp(−αVt+1)/Et exp(−αVt+1)] to a one-standard-deviation (0.7 percent) positive technology shockin the model, with state variables initialized to their nonstochastic steady-state values. Impulse re-sponses are in percentage deviation from steady state. The model-implied stochastic discount factor ismt+1 = β(Ct+1/Ct)
−1[ exp(−αVt+1)/Et exp(−αVt+1)], so the two panels decompose the response of theconditional variance of the SDF to the shock. See text for details.
Et exp(−αVt+1)]. I report nonlinear impulse response functions for each of these two components
in the left- and right-hand panels of Figure 6.37
In response to a positive, one-standard-deviation technology shock, the conditional variance
of the SDF falls about 50 percent. The conditional variance of consumption growth, in the left-
hand panel of Figure 6, falls about 40 percent, while the conditional variance of the Epstein-Zin
term (in the right-hand panel) falls about 10 percent, so the decline in SDF volatility is primarily
driven by the fall in next-period consumption volatility.38 Nevertheless, the Epstein-Zin term
is still extremely important because its average level of volatility is so much higher than that
of consumption growth—about 0.472 vs. 0.00652. The high average level of the Epstein-Zin
term’s volatility is what makes the decline in consumption growth volatility have a quantitatively
important effect on risk premia in the model.
In Figure 7, I investigate what drives the decline in consumption volatility in Figure 6.
37The nonlinear impulse response functions are computed in the same way as in previous figures. I compute theconditional variance of a variable Xt+1 in the model by defining μXt ≡ EtXt+1 and then V X
t ≡ Et(Xt+1 − μXt )2.I append these recursive equations to the rest of the model and solve them nonlinearly along with the other modelvariables as described earlier. Note that the conditional variance V X
t is linearized and not log-linearized aroundthe nonstochastic steady state because it equals zero at that point (when the variance of the technology shockσ2A is set to zero). I report the change in variances V X
t in Figure 6 in percentage terms by dividing the impulse
responses (in levels) by a constant; namely, the variance V Xt solved to fifth order and evaluated with each state
variable at the nonstochastic steady state, but with the variance of the technology shock σ2A set to .0072.
38As with previous figures, the impulse responses in Figure 6 are essentially symmetric for a negative technologyshock. That is, a negative technology shock causes the conditional volatility of the SDF to increase, with a similarmagnitude to (in fact, slightly larger magnitude than) Figure 6.
37
(a) Impulse Responses to .007 Shock εAt in Period 1 (b) Impulse Responses to −.007 Shock εAt in Period 1
no previous shock in period 0
previous shock of .007 in period 0
10 20 30 40 50
-0.4
-0.2
0.2
0.4
percentPrice Dispersion t
no previous shock in period 0
previous shock of .007 in period 0
10 20 30 40 50
-0.4
-0.2
0.2
0.4
percentPrice Dispersion t
0 10 20 30 40 500.0
0.2
0.4
0.6
0.8
1.0percent
ConsumptionCt
10 20 30 40 50
-1.0
-0.8
-0.6
-0.4
-0.2
0.0percent
ConsumptionCt
Figure 7. Comparison of nonlinear impulse response functions for price dispersion Δt and consumption Ct
to a one-standard-deviation (0.7 percent) (a) positive vs. (b) negative technology shock in period 1. Dashedblue lines depict impulse response functions relative to a baseline of no previous shocks, with state variablesinitialized to their nonstochastic steady-state values, as in previous figures. Solid green lines depict impulseresponse functions relative to a different baseline, after a positive 0.7 percent technology shock in period 0.The figure shows that the conditional volatility of consumption is lower after a positive 0.7 percent technologyshock in period 0, particularly in response to negative shocks in period 1. See text for details.
The left column reports nonlinear impulse response functions for a one-standard-deviation (0.7
percent) positive technology shock, while the right column reports the analogous impulse response
functions for a negative one-standard-deviation shock (−0.7 percent). In each panel, two lines
are plotted: The dashed blue line is the standard nonlinear impulse response function computed
in the same way as in previous figures (see footnote 23); that is, the period-by-period difference
between a “one shock” and a “no shock” (baseline) scenario, with all state variables of the model
initialized to their nonstochastic steady-state values. The solid green line in each panel is the
period-by-period difference starting from a different initial point: instead of the nonstochastic
38
steady state, the impulse responses are computed starting from the point immediately after a
positive 0.7 percent technology shock in the previous period. Thus, the solid green lines depict
the difference between a “two shock” and a “one shock” scenario.39
The lower conditional volatility of consumption after a positive technology shock can be seen
clearly in the bottom panels of Figure 7, particularly the bottom-right panel. There, consumption
falls substantially less in response to a negative technology shock (in period 1) if that shock was
preceded by a positive technology shock the period before (in period 0)—that is, the solid green
line does not fall by as much as the dashed blue line. In the bottom-left panel, the response of
consumption to a positive technology shock (in period 1) is fairly similar whether or not there
was a positive technology shock in the previous period (period 0).40
This attenuated behavior of consumption is driven by the response of price dispersion, Δt,
to the technology shock, reported in the top row of Figure 7. Note first how price dispersion
tends to offset the effects of the technology shock: for example, after a negative technology shock
in the right-hand column of Figure 7, price dispersion falls, which tends to increase output, all
else equal (equation 17). Although the net effect of the shock on output and consumption is still
negative (bottom-right panel of Figure 7), the change in price dispersion moderates the effect of
the technology shock. This is true for a positive technology shock in the left-hand column as well.
Importantly, the effect of the technology shock on price dispersion is larger when prices are more
distorted to begin with. For example, in the top-right panel of Figure 7, the solid green line falls
more than the dashed blue line in response to the −0.7 percent shock. In the solid green line
simulation, prices are more distorted to begin with because of the positive technology shock that
hit the economy the period before. Thus, after a positive technology shock, the moderating effects
of price dispersion in the model are greater, causing the volatility of output and consumption to
decrease. In other words, a positive technology shock leads to a lower conditional volatility of
consumption.
39This is computed as follows: in period −1, the state variables of the model are initialized to their nonstochasticsteady-state values. In the “one shock” scenario, εAt is set equal to .007 in period 0, and set equal to 0 from period 1
onward. In the “two shock” scenario, εAt is set equal to .007 in period 0, to .007 in period 1 (for the left-handcolumn of Figure 7, or to −.007 in period 1 for the right-hand column of Figure 7), and then set equal to 0from period 2 onward. The impulse response function is computed as the period-by-period difference between the“two-shock” scenario and the “one shock” scenario, beginning in period 1.
40As in previous figures, the effects are essentially symmetric: if the economy is hit by a negative technologyshock in period 0, then the conditional volatility in Figure 6 increases rather than decreases, and consumption inthe bottom-right panel of Figure 7 is relatively lower after the negative technology shock—i.e., the solid green linelies below the dashed blue line. The lower volatility of consumption growth in the bottom panels of Figure 7 istypical for other sizes of shocks as well.
39
Figure 8. Illustrative graph of price dispersion Δt as a function of monopolistic firms’ time-t real resetprice, p∗t /Pt. Point A denotes firms’ risk-neutral optimal reset price, which is also the optimal resetprice in the linearized model; point B denotes firms’ risk-averse optimal reset price; which leads to aneconomywide equilibrium like point C. See text for details.
Figure 8 provides intuition for why Δt offsets the effect of the technology shock and why
the response of Δt to a shock is larger when Δt itself is larger. First, note that Δt ≥ 1, and
logΔt is a convex function of log(p∗t /Pt) in a neighborhood of the nonstochastic steady state, as
in the figure.41 If the economy is at the nonstochastic steady state and there are no shocks or
uncertainty, then firms find it optimal to set p∗t = Pt in period t. This corresponds to point A in
Figure 8. Even if there is uncertainty, firms still find point A optimal in the linearized version of
the model, because at this point, firms’ expected profits over the lifetime of the price contract are
maximized, as in equation (14). In recessions, firms will lose a bit of profit because output is too
low, and in expansions firms will lose a bit of profit because output (and hence marginal cost) is
too high, but in expectation the firms’ price strikes an optimal balance between these two.
Now, when firms’ owners (the households) are risk averse, as in the nonlinear version of the
model, recessions are more painful than expansions, so it is optimal for firms to put more weight
on generating profit in recessions. This causes the firms’ optimal reset price to be lower than in
the risk-neutral case, so as to generate more output and profit in the event of a recession. Thus,
the risk-averse firms’ optimal reset price in Figure 8 lies at a point like B, to the left of A. Of
course, when all firms act this way, the equilibrium in the economy is at a point like C, where
logΔt > 0 due to firms consistently resetting prices a bit below what the nonstochastic trend rate
41Δt ≥ 1 follows from Jensen’s inequality: because λ > 1 and θ ∈ (0, 1), the function xλ/θ is convex; thus
(∫ 10 x(f)
λ/θdf)θ/λ ≥ ∫ 10 x(f)df ; letting x(f) ≡ (p∗t (f)/Pt)1/(1−λ) gives the result. Moreover, Δt = 1 if and only if
p∗t (f)/Pt = 1 for almost all f . Starting from Δt−1 = 1, p∗t (f)/Pt = 1 attains the minimum value for Δt, implyinglogΔt is convex in log(p∗t (f)/Pt) around 0.
40
of inflation π would imply. Note that at point C, the derivative ∂ logΔt/∂ log(p∗t /Pt) < 0. Thus,
positive technology shocks—which cause firms’ marginal costs to fall and lower p∗t /Pt—raise Δt.
Firms are already setting prices too low from an aggregate efficiency standpoint (because they are
risk-averse), and the positive technology shock exacerbates this inefficiency. Similarly, negative
technology shocks lower Δt. This explains the impulse response functions for Δt in Figure 7, and
why they tend to offset the effects of the technology shock.
To see why this effect is larger when price dispersion is greater, notice that logΔt in Figure 8
is a convex function of log(p∗t /Pt). If Δt increases from point C, this implies moving further up the
curve to the left. At this point, the slope of the curve is more negative. Thus, technology shocks
lead to larger changes in Δt when Δt is larger to begin with, consistent with the results in Figure 7.
It is because of this effect that the conditional volatility of consumption growth and the SDF fall
after a positive technology shock, as in Figure 6. In turn, this conditional heteroskedasticity of
the SDF is what causes risk premia in the model to behave countercyclically.
4.2 Uncertainty Shocks and the Intertemporal Elasticity of Substitution
In the long-run risks literature, such as Bansal and Yaron (2004), it is standard to assume that
the intertemporal elasticity of substitution is substantially greater than unity. There are two
reasons for that calibration: an IES > 1 guarantees that a positive shock to consumption causes
stock prices to rise, and an IES > 1 ensures that an exogenous increase in volatility causes stock
prices to fall.
However, the assumption of an IES > 1 is not strictly necessary for these two criteria to
be satisfied, even in Bansal and Yaron (2004, henceforth BY)’s model. For example, when equity
represents a levered rather than unlevered consumption claim, then equity prices in BY rise in
response to a positive consumption shock if and only if the IES > 1/ν, where ν is the degree of
leverage. For a volatility shock, stock prices respond negatively if and only if the IES > 1/γ,
where γ is the household’s relative risk aversion.42 Since ν, γ > 1, the IES can be less than unity
and still satisfy both of these criteria.
Of course, the model in the present paper differs in many respects from BY. Nevertheless,
a positive shock to technology (and consumption) in the model here causes stock prices to rise,
as can be seen clearly in Figure 2, even though the IES = 1.
42 In Bansal and Yaron (2004), the coefficient A2 for the unlevered consumption claim requires θ < 0, whereθ = (1 − γ)/(1 − 1/ψ) and γ denotes risk aversion and ψ the IES in their paper. For the levered consumptionclaim, however, the coefficient A2,m requires θ/(1− θ) < 0 (see their equation A20), which holds if either θ < 0 orθ > 1. Given γ > 1, then θ > 1 if and only if ψ > 1/γ.
41
To investigate the second criterion—that an increase in volatility causes stock prices to
fall—I extend the model to include exogenous stochastic volatility. In particular, let the standard
deviation of the technology shock each period, σA,t, follow the autoregressive process
where σA = .007, as in Table 1. Following Bansal and Yaron (2004), I calibrate ρσ = 0.98 and
Var(εσt ) = (0.1)2.43
Figure 9 plots the extended model’s nonlinear impulse response functions (computed the
same way as in previous figures) to a positive one-standard-deviation shock to εσt . Volatility
σAt increases to about .0077 on impact and slowly declines back toward its initial level of .007.
Consumption drops about 0.2 percent on impact, as households increase precautionary savings,
and inflation falls gradually by about 0.1 percent in response to the decrease in demand. The
increase in the conditional volatility of consumption increases the volatility of the stochastic
discount factor, which causes a large, 60 bp jump in the equity premium (and a large, 25 bp
increase in the nominal term premium). The large and persistent rise in the equity premium
implies that the equity price must fall dramatically on impact, about 4.5 percent.44 Thus, the
model also satisfies the second criterion described above—that an exogenous increase in volatility
causes stock prices to fall—without the need for an IES > 1, consistent with the discussion above
even though the model differs from BY in many respects.
The exogenous shock to volatility in Figure 9 is related to the literature on “uncertainty
shocks”, such as Bloom (2009), who shows that recessions are typically associated with greater
stock market uncertainty. One interpretation of this correlation is that exogenous increases in
uncertainty lead to recessions, as in Figure 9 (where consumption and output about 0.25 percent
in response to the shock), but it is also possible that the causality runs both ways, so that
recessions lead to higher stock market volatility and uncertainty (e.g., Ludvigson, Ma, and Ng,
2016).
In fact, both of these channels can be seen in the model developed above. In addition to
the impulse responses in Figure 9 to an exogenous uncertainty shock, the model here implies
43Bansal and Yaron (2004) assume a more complicated (square-root rather than logarithmic) process for σA,t
than (38), but the magnitudes in (38) are essentially the same as theirs.44 In order to generate an equity premium of 60 bp in the first period, stock prices must fall by about 0.6 percent
below their second-period value. In order to generate an equity premium in each subsequent period, equity pricesmust continue to rise. This requires a large initial fall in the equity price so that in each subsequent period equityprices can rise in line with the implied equity premium.
42
10 20 30 40 500.0070
0.0072
0.0074
0.0076
0.0078
0.0080Volatility A ,t
10 20 30 40 50
-0.5
-0.4
-0.3
-0.2
-0.1
0.0percent
ConsumptionCt
10 20 30 40 50
-0.5
-0.4
-0.3
-0.2
-0.1
0.0ann. pct.
Inflation t
0 10 20 30 40 500
20
40
60
80
100ann. bp
Equity premium t
e
10 20 30 40 50
-5
-4
-3
-2
-1
0percent
Equity price pte
0 10 20 30 40 500
5
10
15
20
25ann. bp
Nominal term premium t
$ (40)
Figure 9. Nonlinear impulse response functions for volatility σAt , consumption Ct, inflation πt, the equity
premium ψet , equity price pet , and nominal term premium ψ
$(40)t to a positive one-standard-deviation (10
percent, or .0007) volatility shock in the extended model. For clarity, the impulse response for σAt is
graphed as returning to its baseline value of .007 rather than a baseline of 0. See text for details.
43
10 20 30 40 50
-0.30
-0.25
-0.20
-0.15
-0.10
-0.05
0.00percent
Conditional std. dev. of equity price, sdt pt+1e
Figure 10. Nonlinear impulse response function for the one-period-ahead standard deviation of the logequity price, sdt p
et+1, to a one-standard-deviation (0.7 percent) technology shock in the baseline model
with state variables initialized to their steady-state values. See text for details.
that recessions cause stock market uncertainty to increase endogenously.. The intuition for this is
essentially the same as for endogenous conditional heteroskedasticity, above. When the economy is
weak, consumption is low and the household’s stochastic discount factor becomes more sensitive to
subsequent shocks. This drives up the equity premium, as shown in Figure 2, but it also increases
the uncertainty about stock prices, as shown in Figure 10, which plots the nonlinear impulse
response function of the one-step-ahead standard deviation of the log equity price, sdt pet+1 to a
one-standard-deviation (0.7 percent) technology shock, computed the same way as in previous
figures. In response to a positive technology shock, uncertainty about the equity price in the
next period decreases about 0.2 percent—i.e., from about 2.5 percent in Figure 2 to less than 2.3
percent immediately after the shock—a substantial reduction.
4.3 The Financial Accelerator
Traditional models of the financial accelerator (e.g., Bernanke, Gertler, and Gilchrist, 1999; Kiy-
otaki and Moore, 1997; Gertler and Kiyotaki, 2011) allow for the possiblity of default, but ignore
deviations from risk neutrality. The fact that borrowers might default introduces a wedge between
borrowers and lenders that can act as an amplification and propagation mechanism for shocks: for
example, in Kiyotaki and Moore (1997), a negative technology shock reduces the value of capital,
which reduces firms’ collateral; with less collateral, firms must scale back production and output
falls by more than the effect of the technology shock alone. In Bernanke, Gertler, and Gilchrist
(1999, henceforth BGG), a weaker economy implies a higher probability of default, which raises
44
costly state verification costs for financial intermediaries, which in turn leads to a greater spread
between private borrowing rates and the risk-free rate.
The traditional financial accelerator mechanism captures many important features of a
credit crunch and a financial crisis. At the same time, these models are essentially risk-neutral
and abstract from risk premia—that is, in BGG and Kiyotaki and Moore (1997), the spread
between private borrowing rates and the risk-free rate is just the risk-neutral expected loss from
default. Yet an important part of the transmission mechanism in the 2007–08 financial crisis was
the fall in the value of collateral beyond even the risk-neutral probability of default: for example,
risk and liquidity premia rose dramatically even on securities that had little or no connection to
subprime real estate lending (Gorton and Metrick, 2012), and the prices of many mortgage-backed
securities and credit default swaps fell by much more than can be explained by any reasonable
assumption for mortgage default and recovery rates (Stanton and Wallace, 2014). The dramatic
increase in risk premia during the financial crisis caused huge drops in the value of collateral and
historic increases in credit spreads. For the same reasons as in traditional financial accelerator
models, we would expect these repercussions from rising risk premia to be an important part of
the transmission mechanism from financial markets to the real economy.
Recently, some authors have begun to incorporate deviations from risk neutrality into fi-
nancial accelerator models—see, e.g., He and Krishnamurthy (2013), Brunnermeier and Sannikov
(2014)—but those models are idiosyncratic. In contrast, the macroeconomic model I’ve developed
here is canonical and would allow researchers to study the effects of risk premia on collateral val-
ues, credit spreads, and the economy within the well-understood and ubiquitous New Keynesian
DSGE framework. Extending the model to include a financial intermediation sector is beyond
the scope of the present paper, but would be a worthy topic for future research.
5. Conclusions
This paper shows that a simple, textbook New Keyneisan model with Epstein-Zin preferences
is consistent with a wide variety of asset pricing facts, such as the equity premium, real and
nominal term premium, and credit spread. The basic New Keynesian model produces the correct
correlations between the macroeconomy, real and nominal interest rates, and dividends, while
Epstein-Zin preferences allow the model to match the overall size and variability of the risk
premia on assets.
45
For simplicity, I have kept the model as simple as possible and used Epstein-Zin preferences
with a high degree of risk aversion to match the size of risk premia in the data. This degree of
risk aversion can be lowered dramatically by augmenting the model to include additional sources
of risk. For example, Barillas, Hansen, and Sargent (2009) show there is a formal mathematical
equivalence between Epstein-Zin preferences with a high degree of risk aversion and those same
preferences with a lower degree of risk aversion but a moderate level of uncertainty about the
structure and parameters of the model. Similarly, Schmidt (2015), following Constantinides
and Duffie (1996), shows that Epstein-Zin preferences with a high degree of risk aversion in a
representative agent model are formally equivalent to those same preferences with a lower degree
of risk aversion and heterogeneous agents who face uninsurable idiosyncratic risks. One can
similarly reduce the required risk aversion in the model by adding long-run risks (e.g., Bansal
and Yaron, 2004), rare disasters (e.g., Rietz, 1988; Barro, 2006), or parameter uncertainty (e.g.,
Weitzman, 2007). All of these approaches address the fact that the quantity of risk faced by
a representative household in a textbook New Keynesian model with i.i.d. normally-distributed
shocks is not very large. The point of the present paper is not to incorporate all of these additional
features, but rather to serve as a “proof of concept” that the asset pricing data can be matched
within the standard New Keynesian modeling framework.
The simple, structural model I develop provides a unified and intuitive framework for think-
ing about asset prices and asset pricing puzzles. Rather than studying each puzzle in isolation,
the model here provides a reasonable description of the behavior of all of the major asset classes.
In addition, structural models have the well-known advantage of being more robust to structural
breaks and novel policy interventions, such as those observed during the recent global financial cri-
sis and European sovereign debt crisis. The model developed here can potentially provide insight
in these situations, when more traditional, reduced-form models are largely uninformative.
Finally, by showing how a standard macroeconomic model can be made consistent with
the behavior of risk premia in financial markets, the present paper opens the door to studying
the feedback between those risk premia and the economy within the standard macreconomic
modeling framework in use at central banks and other policy institutions. As evidenced by the
recent financial crises, this feedback from asset prices to the economy and back again can be very
important. In the simple, stylized model of the present paper, asset prices have no feedback effects
on the real economy, for simplicity, but it would be very interesting to combine the asset-pricing
framework of the present paper with a macroeconomic model that includes a financial accelerator
46
along the lines of Bernanke, Gertler, and Gilchrist (1999), Kiyotaki and Moore (1997), Gertler
and Kiyotaki (2011), and others. In general, these models abstract from risk aversion and risk
premia and focus instead on the effect of agency problems and collateral constraints on lending
and investment. In a combined framework, shocks that cause the economy to deteriorate would
lead to an increase in risk premia and a concomitant fall in asset prices, further amplifying the
collateral constraint on firms and financial intermediaries. This channel appears to have been an
important amplification and propagation mechanism in recent crises.
47
Appendix A: Model Equations
We can write the equations of the macroeconomic model in Section 2 in recursive form as follows. (Equa-tions for equity and debt are essentially the same as in Section 3 and are not reproduced here.)
Equations (A1)–(A2) break the generalized value function into two equations to correspond to thesyntax of Perturbation AIM and other rational expectations equation solvers, which typically require themodel to be written as a system of equations in a form similar to EtF (Xt−1, Xt, Xt+1; εt) = 0.
Equations (A5)–(A7) represent monopolistic firms’ optimal price-setting equations. The exponenton (p∗t /Pt) in (A5) follows from substituting out yt+j(f) in equation (14) in the main text, and is dueto the presence of firm-specific capital stocks. Equations (A6)–(A7) are recursive versions of the infinitesums in the numerator and denominator of (14).
The other equations above follow in a straightforward manner from the equations in the main text.
48
Although capital stocks in the model above are fixed, the model nevertheless has a balanced growthpath along which all variables are either constant or grow at constant rates if technology At itself growsat a constant rate. Along the balanced growth path, each of the variables Yt, Ct, wt, Yt, z
nt , and z
dt grow
at the same rate as At. If we divide each of these variables through by At, the ratios have a nonstochasticsteady state. Moreover, after a shock to At, these ratios converge back to their pre-shock levels. Thus,the nonstochastic steady state of these ratios constitutes a stable point around which we can approximatethe model.
I thus transform the model by dividing each of the above variables by the level of technology At,and transform the value function Vt by defining Vt ≡ Vt − logAt. The transformed model then has anonstochastic steady state around which I can compute an nth-order approximate solution as describedin the text. These solutions are highly accurate in a neighborhood of the steady state, and becomeincreasingly accurate over larger regions of the state space as the order of approximation n becomes large(see Swanson, Anderson, and Levin, 2006, for details and discussion).
Appendix B: Impulse Responses to a Negative Technology Shock
The nonlinear impulse response functions graphed in Figures 1 through 5 can be asymmetric for positiveand negative shocks, because they are nonlinear. In practice, the New Keynesian model presented inthe main text does not produce impulse response functions that are very asymmetric. This can be seen,for example, in Figure/ B1, which reproduces Figure 1 from the main text for the case of a negativeone-standard-deviation (−.007) shock to technology At. The nonlinear impulse response functions inFigure B1 are computed in exactly the same was as in Figures 1 through 6, except that the shock has theopposite sign. Overall, the responses in Figure B1 are close to being symmetric counterparts to Figure 1.
Figure B2 presents nonlinear impulse response functions for the equity price pet , equity premiumψe
t , real long-term bond price p(40)t , and term premium ψ
(40)t on the real long-term bond to the negative
one-standard-deviation technology shock. The impulse responses are again close to being the symmetriccounterparts to the nonlinear impulse responses functions in Figures 2 and 3, although the asset priceand risk premium responses are slightly larger in magnitude for the negative shock than they are for thepositive shock. For example, the equity premium increases by about 70 bp after the negative technologyshock here, while it fell by about 62 bp after the positive shock in Figure 2. Similarly, the equity pricefalls by about 2.75 percent after the negative technology shock here, but rose by about 2.5 percent afterthe positive shock in Figure 2.
Figure B3 repeats the analysis for the nominal long-term bond price p$(40)t , the term premium ψ
$(40)t
on the nominal long-term bond, defaultable bond price pdt , and credit spread idt − ict . As with Figure B2,the responses here are essentially symmetric to their counterparts in Figures 4 and 5, while also beingsomewhat larger in magnitude.
49
1st-order solution
5th-order solution
10 20 30 40 50
-1.0
-0.6
-0.8
-0.4
-0.2
0.0percent
ConsumptionCt
0 10 20 30 40 50 0
0.2
0.4
0.6
0.8
1.0ann. pct.
Inflation t
0 10 20 30 40 500.0
0.2
0.4
0.6
0.8ann. pct.
Short-term nominal interest rate it
10 20 30 40 50
-0.5
-0.4
-0.3
-0.2
-0.1
0.0ann. pct.
Short-term reai lnterest rate rt
10 20 30 40 50
-0.4
-0.2
0.0
0.2
0.4
percentLabor Lt
10 20 30 40 50
-1.0
-0.8
-0.6
-0.4
-0.2
0percent
Price Dispersion t
Figure B1. First-order (dashed red lines) and fifth-order (solid blue lines) impulse response functions forconsumption Ct, inflation πt, short-term nominal interest rate it, short-term real interest rate rt, laborLt, and price dispersion Δt to a one-standard-deviation negative (−0.7 percent) technology shock in themodel. See Figure 1 for comparison and text for details.
50
10 20 30 40 50
-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
0percent
Equity price pte
0 10 20 30 40 500
20
40
60
80ann. bp
Equity premium t
e
0 10 20 30 40 50 0
0.2
0.4
0.6
0.8
1.0percent
Real long-term bond price pt (41)
0 10 20 30 40 500
2
4
6
8
10ann. bp
Real term premium t
(41)
Figure B2. Nonlinear impulse response functions for the equity price pet , equity premium ψet , real long-
term bond price p(40)t and real term premium ψ
(40)t to a one-standard-deviation negative (−0.7 percent)
technology shock in the model, with state variables initialized to their nonstochastic steady state values.See Figures 2–3 for comparison and text for details.
51
10 20 30 40 50
-2.5
-2.0
-1.5
-1.0
-0.5
0percent
Nominal long-term bond price pt$(41)
0 10 20 30 40 500
2
4
6
8
10
12
14
ann. bpNominal term premium t
$(41)
10 20 30 40 50
-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
0percent
Nominal defaultable long-term bond price ptd
0 10 20 30 40 500
2
4
6
8
10ann. bp
Credit spread itd - itc
Figure B3. Nonlinear impulse response functions for the nominal long-term bond price p$(40)t , nominal
bond term premium ψ$(40)t , defaultable long-term bond price pdt , and credit spread idt − ict to a one-
standard-deviation negative (−0.7 percent) technology shock in the model, with state variables initializedto their nonstochastic steady state values. See Figures 4–5 for comparison and text for details.
52
References
Abel, Andrew (1999). “Risk Premia and Term Premia in General Equilibrium,” Journal of Monetary
Economics 43, 3–33.
Adrian, Tobias, Erkko Etula, and Tyler Muir (2014). “Financial Intermediaries and the Cross-
Section of Asset Returns,” Journal of Finance, forthcoming.
Altig, David, Lawrence Christiano, Martin Eichenbaum, and Jesper Linde (2011). “Firm-
Specific Capital, Nominal Rigidities, and the Business Cycle,” Review of Economic Dynamics 14,
225–247.
Altonji, Joseph (1986). “Intertemporal Substitution in Labor Supply: Evidence from Micro Data,”
Journal of Political Economy 94, S176–S215.
Andreasen, Martin (2012). “An Estimated DSGE Model: Explaining Variation in Nominal Term
Premia, Real Term Premia, and Inflation Risk Premia,” European Economic Review 56, 1656–1674.
Ang, Andrew, and Maxim Ulrich (2012). “Nominal Bonds, Real Bonds, and Equity,” unpublished
manuscript, Columbia Business School.
Aruoba, S. Boragan, Jesus Fernandez-Villaverde, and Juan Rubio-Ramırez (2006). “Compar-
ing Solution Methods for Dynamic Equilibrium Economies,” Journal of Economic Dynamics and
Control 30, 2477–2508.
Backus, David, Allan Gregory, and Stanley Zin (1989). “Risk Premiums in the Term Structure,”
Journal of Monetary Economics 24, 371–399.
Baele, Lieven, Geert Bekaert, and Koen Inghelbrecht (2010). “The Determinants of Stock and
Bond Comovements,” Review of Financial Studies 23, 2374–2428.
Bansal, Ravi, and Amir Yaron (2004). “Risks for the Long Run: A Potential Resolution of Asset-
Pricing Puzzles,” Journal of Finance 59, 1481–1509.
Barillas, Francisco, Lars Hansen, and Thomas Sargent (2009). “Doubts or Variability?” Journal
of Economic Theory 144, 2388–2418.
Barro, Robert (2006). “Rare Disasters and Asset Markets in the Twentieth Century,” Quarterly Jour-
nal of Economics 121, 823–866.
Bekaert, Geert, Seonghoon Cho, and Antonio Moreno (2010). “New Keynesian Macroeconomics
and the Term Structure,” Journal of Money, Credit and Banking 42, 33–62.
Bekaert, Geert, Eric Engstrom, and Steven Grenadier (2010). “Stock and Bond Returns with
Moody Investors,” Journal of Empirical Finance 17, 867–894.
Bernanke, Ben, Mark Gertler, and Simon Gilchrist (1999). “The Financial Accelerator in a
Quantitative Business Cycle Framework,” Handbook of Macroeconomics 1, 1341–1393.
Bhamra, Harjoat, Lars-Alexander Kuehn, and Ilya Strebulaev (2010). “The Levered Equity
Risk Premium and Credit Spreads: A Unified Framework,” Review of Financial Studies 23, 645–703.
Bloom, Nicholas (2009). “The Impact of Uncertainty Shocks,” Econometrica 77, 623–685.
Boldrin, Michele, Lawrence Christiano, and Jonas Fisher (2001). “Habit Persistence, Asset
Returns, and the Business Cycle,” American Economic Review 91, 149–166.
Brunnermeier, Markus, and Yuliy Sannikov (2014). “A Macroeconomic Model with a Financial
Sector,” American Economic Review 104, 379–421.
Calvo, Guillermo (1983). “Staggered Prices in a Utility-Maximizing Framework,” Journal of Monetary
Economics 12, 383–398.
Campanale, Claudio, Rui Castro, and Gian Luca Clementi (2010). “Asset Pricing in a Production
Economy with Chew-Dekel Preferences,” Review of Economic Dynamics 13, 379–402.
53
Campbell, John (1999). “Asset Prices, Consumption, and the Business Cycle,” Handbook of Macroe-
conomics 1, ed. John Taylor and Michael Woodford, Elsevier, 1231–1303.
Campbell, John, and John Cochrane (1999). “By Force of Habit: A Consumption-Based Explanation
of Aggregate Stock Market Behavior,” Journal of Political Economy 107, 205–251.
Campbell, John, Carolin Pflueger, and Luis Viceira (2018). “Macroeconomic Drivers of Bond
and Equity Risks,” unpublished manuscript, Harvard University.
Campbell, John, Adi Sundaram, and Luis Viceira (2013). “Inflation Bets or Deflation Hedges? The
Changing Risks of Nominal Bonds,” unpublished manuscript, Harvard University.
Chen, Hui (2010). “Macroeconomic Conditions and the Puzzles of Credit Spreads and Capital Struc-
ture,” Journal of Finance 65, 2171–2212.
Chen, Long, Pierre Collin-Dufresne, and Robert Goldstein (2009). “On the Relation Between
the Credit Spread Puzzle and the Equity Premium Puzzle,” Review of Financial Studies 22, 3367–
3409.
Christiano, Lawrence, Martin Eichenbaum, and Charles Evans (2005). “Nominal Rigidities and
the Dynamic Effects of a Shock to Monetary Policy,” Journal of Political Economy 113, 1–45.
Clarida, Richard, Jordi Gali, and Mark Gertler (1999). “The Science of Monetary Policy: A
New Keynesian Perspective,” Journal of Economic Literature 37, 1661–1707.
Cochrane, John (2008). “Financial Markets and the Real Economy,” Handbook of the Equity Risk
Premium, 237–326.
Cochrane, John, and Monika Piazzesi (2005). “Bond Risk Premia,” American Economic Review 95,
138–160.
Constantinides, George, and Darrell Duffie (1996). “Asset Pricing with Heterogeneous Con-
sumers,” Journal of Political Economy 104, 219–240.
Cooper, Ilan, and Richard Priestley (2008). “Time-Varying Risk Premiums and the Output Gap,”
Review of Financial Studies 22, 2801–2833.
Dai, Qiang, and Kenneth Singleton (2003). “Fixed-Income Pricing,” Handbook of the Economics of
Finance 1B, 1207–1246.
Del Negro, Marco, Marc Giannoni, and Frank Schorfheide (2015). “Inflation in the Great
Recession and New Keynesian Models,” American Economic Journal: Macroeconomics 7, 168–196.
Den Haan, Wouter (1995). “The Term Structure of Interest Rates in Real and Monetary Economies,”
Journal of Economic Dynamics and Control 19, 909–940.
Dew-Becker, Ian (2012). “A Model of Time-Varying Risk Premia with Habits and Production,” un-
published manuscript, Fuqua School of Business, Duke University.
Donaldson, John, Thore Johnsen, and Rajnish Mehra (1990). “On the Term Structure of Interest
Rates,” Journal of Economic Dynamics and Control 14, 571–596.
Duffie, Darrel, and David Lando (2001). “Term Structures of Credit Spreads with Incomplete